2 * Block driver for the QCOW version 2 format
4 * Copyright (c) 2004-2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
30 #include "block/block_int.h"
31 #include "block/qdict.h"
32 #include "sysemu/block-backend.h"
33 #include "qemu/module.h"
35 #include "qemu/error-report.h"
36 #include "qapi/error.h"
37 #include "qapi/qapi-events-block-core.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qstring.h"
41 #include "qemu/option_int.h"
42 #include "qemu/cutils.h"
43 #include "qemu/bswap.h"
44 #include "qapi/qobject-input-visitor.h"
45 #include "qapi/qapi-visit-block-core.h"
47 #include "block/thread-pool.h"
50 Differences with QCOW:
52 - Support for multiple incremental snapshots.
53 - Memory management by reference counts.
54 - Clusters which have a reference count of one have the bit
55 QCOW_OFLAG_COPIED to optimize write performance.
56 - Size of compressed clusters is stored in sectors to reduce bit usage
57 in the cluster offsets.
58 - Support for storing additional data (such as the VM state) in the
60 - If a backing store is used, the cluster size is not constrained
61 (could be backported to QCOW).
62 - L2 tables have always a size of one cluster.
69 } QEMU_PACKED QCowExtension
;
71 #define QCOW2_EXT_MAGIC_END 0
72 #define QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
73 #define QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
74 #define QCOW2_EXT_MAGIC_CRYPTO_HEADER 0x0537be77
75 #define QCOW2_EXT_MAGIC_BITMAPS 0x23852875
76 #define QCOW2_EXT_MAGIC_DATA_FILE 0x44415441
78 static int coroutine_fn
79 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
80 uint64_t file_cluster_offset
,
85 static int qcow2_probe(const uint8_t *buf
, int buf_size
, const char *filename
)
87 const QCowHeader
*cow_header
= (const void *)buf
;
89 if (buf_size
>= sizeof(QCowHeader
) &&
90 be32_to_cpu(cow_header
->magic
) == QCOW_MAGIC
&&
91 be32_to_cpu(cow_header
->version
) >= 2)
98 static ssize_t
qcow2_crypto_hdr_read_func(QCryptoBlock
*block
, size_t offset
,
99 uint8_t *buf
, size_t buflen
,
100 void *opaque
, Error
**errp
)
102 BlockDriverState
*bs
= opaque
;
103 BDRVQcow2State
*s
= bs
->opaque
;
106 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
107 error_setg(errp
, "Request for data outside of extension header");
111 ret
= bdrv_pread(bs
->file
,
112 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
114 error_setg_errno(errp
, -ret
, "Could not read encryption header");
121 static ssize_t
qcow2_crypto_hdr_init_func(QCryptoBlock
*block
, size_t headerlen
,
122 void *opaque
, Error
**errp
)
124 BlockDriverState
*bs
= opaque
;
125 BDRVQcow2State
*s
= bs
->opaque
;
129 ret
= qcow2_alloc_clusters(bs
, headerlen
);
131 error_setg_errno(errp
, -ret
,
132 "Cannot allocate cluster for LUKS header size %zu",
137 s
->crypto_header
.length
= headerlen
;
138 s
->crypto_header
.offset
= ret
;
140 /* Zero fill remaining space in cluster so it has predictable
141 * content in case of future spec changes */
142 clusterlen
= size_to_clusters(s
, headerlen
) * s
->cluster_size
;
143 assert(qcow2_pre_write_overlap_check(bs
, 0, ret
, clusterlen
, false) == 0);
144 ret
= bdrv_pwrite_zeroes(bs
->file
,
146 clusterlen
- headerlen
, 0);
148 error_setg_errno(errp
, -ret
, "Could not zero fill encryption header");
156 static ssize_t
qcow2_crypto_hdr_write_func(QCryptoBlock
*block
, size_t offset
,
157 const uint8_t *buf
, size_t buflen
,
158 void *opaque
, Error
**errp
)
160 BlockDriverState
*bs
= opaque
;
161 BDRVQcow2State
*s
= bs
->opaque
;
164 if ((offset
+ buflen
) > s
->crypto_header
.length
) {
165 error_setg(errp
, "Request for data outside of extension header");
169 ret
= bdrv_pwrite(bs
->file
,
170 s
->crypto_header
.offset
+ offset
, buf
, buflen
);
172 error_setg_errno(errp
, -ret
, "Could not read encryption header");
180 * read qcow2 extension and fill bs
181 * start reading from start_offset
182 * finish reading upon magic of value 0 or when end_offset reached
183 * unknown magic is skipped (future extension this version knows nothing about)
184 * return 0 upon success, non-0 otherwise
186 static int qcow2_read_extensions(BlockDriverState
*bs
, uint64_t start_offset
,
187 uint64_t end_offset
, void **p_feature_table
,
188 int flags
, bool *need_update_header
,
191 BDRVQcow2State
*s
= bs
->opaque
;
195 Qcow2BitmapHeaderExt bitmaps_ext
;
197 if (need_update_header
!= NULL
) {
198 *need_update_header
= false;
202 printf("qcow2_read_extensions: start=%ld end=%ld\n", start_offset
, end_offset
);
204 offset
= start_offset
;
205 while (offset
< end_offset
) {
209 if (offset
> s
->cluster_size
)
210 printf("qcow2_read_extension: suspicious offset %lu\n", offset
);
212 printf("attempting to read extended header in offset %lu\n", offset
);
215 ret
= bdrv_pread(bs
->file
, offset
, &ext
, sizeof(ext
));
217 error_setg_errno(errp
, -ret
, "qcow2_read_extension: ERROR: "
218 "pread fail from offset %" PRIu64
, offset
);
221 ext
.magic
= be32_to_cpu(ext
.magic
);
222 ext
.len
= be32_to_cpu(ext
.len
);
223 offset
+= sizeof(ext
);
225 printf("ext.magic = 0x%x\n", ext
.magic
);
227 if (offset
> end_offset
|| ext
.len
> end_offset
- offset
) {
228 error_setg(errp
, "Header extension too large");
233 case QCOW2_EXT_MAGIC_END
:
236 case QCOW2_EXT_MAGIC_BACKING_FORMAT
:
237 if (ext
.len
>= sizeof(bs
->backing_format
)) {
238 error_setg(errp
, "ERROR: ext_backing_format: len=%" PRIu32
239 " too large (>=%zu)", ext
.len
,
240 sizeof(bs
->backing_format
));
243 ret
= bdrv_pread(bs
->file
, offset
, bs
->backing_format
, ext
.len
);
245 error_setg_errno(errp
, -ret
, "ERROR: ext_backing_format: "
246 "Could not read format name");
249 bs
->backing_format
[ext
.len
] = '\0';
250 s
->image_backing_format
= g_strdup(bs
->backing_format
);
252 printf("Qcow2: Got format extension %s\n", bs
->backing_format
);
256 case QCOW2_EXT_MAGIC_FEATURE_TABLE
:
257 if (p_feature_table
!= NULL
) {
258 void* feature_table
= g_malloc0(ext
.len
+ 2 * sizeof(Qcow2Feature
));
259 ret
= bdrv_pread(bs
->file
, offset
, feature_table
, ext
.len
);
261 error_setg_errno(errp
, -ret
, "ERROR: ext_feature_table: "
262 "Could not read table");
266 *p_feature_table
= feature_table
;
270 case QCOW2_EXT_MAGIC_CRYPTO_HEADER
: {
271 unsigned int cflags
= 0;
272 if (s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
273 error_setg(errp
, "CRYPTO header extension only "
274 "expected with LUKS encryption method");
277 if (ext
.len
!= sizeof(Qcow2CryptoHeaderExtension
)) {
278 error_setg(errp
, "CRYPTO header extension size %u, "
279 "but expected size %zu", ext
.len
,
280 sizeof(Qcow2CryptoHeaderExtension
));
284 ret
= bdrv_pread(bs
->file
, offset
, &s
->crypto_header
, ext
.len
);
286 error_setg_errno(errp
, -ret
,
287 "Unable to read CRYPTO header extension");
290 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
291 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
293 if ((s
->crypto_header
.offset
% s
->cluster_size
) != 0) {
294 error_setg(errp
, "Encryption header offset '%" PRIu64
"' is "
295 "not a multiple of cluster size '%u'",
296 s
->crypto_header
.offset
, s
->cluster_size
);
300 if (flags
& BDRV_O_NO_IO
) {
301 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
303 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
304 qcow2_crypto_hdr_read_func
,
305 bs
, cflags
, 1, errp
);
311 case QCOW2_EXT_MAGIC_BITMAPS
:
312 if (ext
.len
!= sizeof(bitmaps_ext
)) {
313 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
314 "Invalid extension length");
318 if (!(s
->autoclear_features
& QCOW2_AUTOCLEAR_BITMAPS
)) {
319 if (s
->qcow_version
< 3) {
320 /* Let's be a bit more specific */
321 warn_report("This qcow2 v2 image contains bitmaps, but "
322 "they may have been modified by a program "
323 "without persistent bitmap support; so now "
324 "they must all be considered inconsistent");
326 warn_report("a program lacking bitmap support "
327 "modified this file, so all bitmaps are now "
328 "considered inconsistent");
330 error_printf("Some clusters may be leaked, "
331 "run 'qemu-img check -r' on the image "
333 if (need_update_header
!= NULL
) {
334 /* Updating is needed to drop invalid bitmap extension. */
335 *need_update_header
= true;
340 ret
= bdrv_pread(bs
->file
, offset
, &bitmaps_ext
, ext
.len
);
342 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
343 "Could not read ext header");
347 if (bitmaps_ext
.reserved32
!= 0) {
348 error_setg_errno(errp
, -ret
, "bitmaps_ext: "
349 "Reserved field is not zero");
353 bitmaps_ext
.nb_bitmaps
= be32_to_cpu(bitmaps_ext
.nb_bitmaps
);
354 bitmaps_ext
.bitmap_directory_size
=
355 be64_to_cpu(bitmaps_ext
.bitmap_directory_size
);
356 bitmaps_ext
.bitmap_directory_offset
=
357 be64_to_cpu(bitmaps_ext
.bitmap_directory_offset
);
359 if (bitmaps_ext
.nb_bitmaps
> QCOW2_MAX_BITMAPS
) {
361 "bitmaps_ext: Image has %" PRIu32
" bitmaps, "
362 "exceeding the QEMU supported maximum of %d",
363 bitmaps_ext
.nb_bitmaps
, QCOW2_MAX_BITMAPS
);
367 if (bitmaps_ext
.nb_bitmaps
== 0) {
368 error_setg(errp
, "found bitmaps extension with zero bitmaps");
372 if (bitmaps_ext
.bitmap_directory_offset
& (s
->cluster_size
- 1)) {
373 error_setg(errp
, "bitmaps_ext: "
374 "invalid bitmap directory offset");
378 if (bitmaps_ext
.bitmap_directory_size
>
379 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
) {
380 error_setg(errp
, "bitmaps_ext: "
381 "bitmap directory size (%" PRIu64
") exceeds "
382 "the maximum supported size (%d)",
383 bitmaps_ext
.bitmap_directory_size
,
384 QCOW2_MAX_BITMAP_DIRECTORY_SIZE
);
388 s
->nb_bitmaps
= bitmaps_ext
.nb_bitmaps
;
389 s
->bitmap_directory_offset
=
390 bitmaps_ext
.bitmap_directory_offset
;
391 s
->bitmap_directory_size
=
392 bitmaps_ext
.bitmap_directory_size
;
395 printf("Qcow2: Got bitmaps extension: "
396 "offset=%" PRIu64
" nb_bitmaps=%" PRIu32
"\n",
397 s
->bitmap_directory_offset
, s
->nb_bitmaps
);
401 case QCOW2_EXT_MAGIC_DATA_FILE
:
403 s
->image_data_file
= g_malloc0(ext
.len
+ 1);
404 ret
= bdrv_pread(bs
->file
, offset
, s
->image_data_file
, ext
.len
);
406 error_setg_errno(errp
, -ret
,
407 "ERROR: Could not read data file name");
411 printf("Qcow2: Got external data file %s\n", s
->image_data_file
);
417 /* unknown magic - save it in case we need to rewrite the header */
418 /* If you add a new feature, make sure to also update the fast
419 * path of qcow2_make_empty() to deal with it. */
421 Qcow2UnknownHeaderExtension
*uext
;
423 uext
= g_malloc0(sizeof(*uext
) + ext
.len
);
424 uext
->magic
= ext
.magic
;
426 QLIST_INSERT_HEAD(&s
->unknown_header_ext
, uext
, next
);
428 ret
= bdrv_pread(bs
->file
, offset
, uext
->data
, uext
->len
);
430 error_setg_errno(errp
, -ret
, "ERROR: unknown extension: "
431 "Could not read data");
438 offset
+= ((ext
.len
+ 7) & ~7);
444 static void cleanup_unknown_header_ext(BlockDriverState
*bs
)
446 BDRVQcow2State
*s
= bs
->opaque
;
447 Qcow2UnknownHeaderExtension
*uext
, *next
;
449 QLIST_FOREACH_SAFE(uext
, &s
->unknown_header_ext
, next
, next
) {
450 QLIST_REMOVE(uext
, next
);
455 static void report_unsupported_feature(Error
**errp
, Qcow2Feature
*table
,
458 char *features
= g_strdup("");
461 while (table
&& table
->name
[0] != '\0') {
462 if (table
->type
== QCOW2_FEAT_TYPE_INCOMPATIBLE
) {
463 if (mask
& (1ULL << table
->bit
)) {
465 features
= g_strdup_printf("%s%s%.46s", old
, *old
? ", " : "",
468 mask
&= ~(1ULL << table
->bit
);
476 features
= g_strdup_printf("%s%sUnknown incompatible feature: %" PRIx64
,
477 old
, *old
? ", " : "", mask
);
481 error_setg(errp
, "Unsupported qcow2 feature(s): %s", features
);
486 * Sets the dirty bit and flushes afterwards if necessary.
488 * The incompatible_features bit is only set if the image file header was
489 * updated successfully. Therefore it is not required to check the return
490 * value of this function.
492 int qcow2_mark_dirty(BlockDriverState
*bs
)
494 BDRVQcow2State
*s
= bs
->opaque
;
498 assert(s
->qcow_version
>= 3);
500 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
501 return 0; /* already dirty */
504 val
= cpu_to_be64(s
->incompatible_features
| QCOW2_INCOMPAT_DIRTY
);
505 ret
= bdrv_pwrite(bs
->file
, offsetof(QCowHeader
, incompatible_features
),
510 ret
= bdrv_flush(bs
->file
->bs
);
515 /* Only treat image as dirty if the header was updated successfully */
516 s
->incompatible_features
|= QCOW2_INCOMPAT_DIRTY
;
521 * Clears the dirty bit and flushes before if necessary. Only call this
522 * function when there are no pending requests, it does not guard against
523 * concurrent requests dirtying the image.
525 static int qcow2_mark_clean(BlockDriverState
*bs
)
527 BDRVQcow2State
*s
= bs
->opaque
;
529 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
532 s
->incompatible_features
&= ~QCOW2_INCOMPAT_DIRTY
;
534 ret
= qcow2_flush_caches(bs
);
539 return qcow2_update_header(bs
);
545 * Marks the image as corrupt.
547 int qcow2_mark_corrupt(BlockDriverState
*bs
)
549 BDRVQcow2State
*s
= bs
->opaque
;
551 s
->incompatible_features
|= QCOW2_INCOMPAT_CORRUPT
;
552 return qcow2_update_header(bs
);
556 * Marks the image as consistent, i.e., unsets the corrupt bit, and flushes
557 * before if necessary.
559 int qcow2_mark_consistent(BlockDriverState
*bs
)
561 BDRVQcow2State
*s
= bs
->opaque
;
563 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
564 int ret
= qcow2_flush_caches(bs
);
569 s
->incompatible_features
&= ~QCOW2_INCOMPAT_CORRUPT
;
570 return qcow2_update_header(bs
);
575 static int coroutine_fn
qcow2_co_check_locked(BlockDriverState
*bs
,
576 BdrvCheckResult
*result
,
579 int ret
= qcow2_check_refcounts(bs
, result
, fix
);
584 if (fix
&& result
->check_errors
== 0 && result
->corruptions
== 0) {
585 ret
= qcow2_mark_clean(bs
);
589 return qcow2_mark_consistent(bs
);
594 static int coroutine_fn
qcow2_co_check(BlockDriverState
*bs
,
595 BdrvCheckResult
*result
,
598 BDRVQcow2State
*s
= bs
->opaque
;
601 qemu_co_mutex_lock(&s
->lock
);
602 ret
= qcow2_co_check_locked(bs
, result
, fix
);
603 qemu_co_mutex_unlock(&s
->lock
);
607 int qcow2_validate_table(BlockDriverState
*bs
, uint64_t offset
,
608 uint64_t entries
, size_t entry_len
,
609 int64_t max_size_bytes
, const char *table_name
,
612 BDRVQcow2State
*s
= bs
->opaque
;
614 if (entries
> max_size_bytes
/ entry_len
) {
615 error_setg(errp
, "%s too large", table_name
);
619 /* Use signed INT64_MAX as the maximum even for uint64_t header fields,
620 * because values will be passed to qemu functions taking int64_t. */
621 if ((INT64_MAX
- entries
* entry_len
< offset
) ||
622 (offset_into_cluster(s
, offset
) != 0)) {
623 error_setg(errp
, "%s offset invalid", table_name
);
630 static QemuOptsList qcow2_runtime_opts
= {
632 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_runtime_opts
.head
),
635 .name
= QCOW2_OPT_LAZY_REFCOUNTS
,
636 .type
= QEMU_OPT_BOOL
,
637 .help
= "Postpone refcount updates",
640 .name
= QCOW2_OPT_DISCARD_REQUEST
,
641 .type
= QEMU_OPT_BOOL
,
642 .help
= "Pass guest discard requests to the layer below",
645 .name
= QCOW2_OPT_DISCARD_SNAPSHOT
,
646 .type
= QEMU_OPT_BOOL
,
647 .help
= "Generate discard requests when snapshot related space "
651 .name
= QCOW2_OPT_DISCARD_OTHER
,
652 .type
= QEMU_OPT_BOOL
,
653 .help
= "Generate discard requests when other clusters are freed",
656 .name
= QCOW2_OPT_OVERLAP
,
657 .type
= QEMU_OPT_STRING
,
658 .help
= "Selects which overlap checks to perform from a range of "
659 "templates (none, constant, cached, all)",
662 .name
= QCOW2_OPT_OVERLAP_TEMPLATE
,
663 .type
= QEMU_OPT_STRING
,
664 .help
= "Selects which overlap checks to perform from a range of "
665 "templates (none, constant, cached, all)",
668 .name
= QCOW2_OPT_OVERLAP_MAIN_HEADER
,
669 .type
= QEMU_OPT_BOOL
,
670 .help
= "Check for unintended writes into the main qcow2 header",
673 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L1
,
674 .type
= QEMU_OPT_BOOL
,
675 .help
= "Check for unintended writes into the active L1 table",
678 .name
= QCOW2_OPT_OVERLAP_ACTIVE_L2
,
679 .type
= QEMU_OPT_BOOL
,
680 .help
= "Check for unintended writes into an active L2 table",
683 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
684 .type
= QEMU_OPT_BOOL
,
685 .help
= "Check for unintended writes into the refcount table",
688 .name
= QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
689 .type
= QEMU_OPT_BOOL
,
690 .help
= "Check for unintended writes into a refcount block",
693 .name
= QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
694 .type
= QEMU_OPT_BOOL
,
695 .help
= "Check for unintended writes into the snapshot table",
698 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L1
,
699 .type
= QEMU_OPT_BOOL
,
700 .help
= "Check for unintended writes into an inactive L1 table",
703 .name
= QCOW2_OPT_OVERLAP_INACTIVE_L2
,
704 .type
= QEMU_OPT_BOOL
,
705 .help
= "Check for unintended writes into an inactive L2 table",
708 .name
= QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
709 .type
= QEMU_OPT_BOOL
,
710 .help
= "Check for unintended writes into the bitmap directory",
713 .name
= QCOW2_OPT_CACHE_SIZE
,
714 .type
= QEMU_OPT_SIZE
,
715 .help
= "Maximum combined metadata (L2 tables and refcount blocks) "
719 .name
= QCOW2_OPT_L2_CACHE_SIZE
,
720 .type
= QEMU_OPT_SIZE
,
721 .help
= "Maximum L2 table cache size",
724 .name
= QCOW2_OPT_L2_CACHE_ENTRY_SIZE
,
725 .type
= QEMU_OPT_SIZE
,
726 .help
= "Size of each entry in the L2 cache",
729 .name
= QCOW2_OPT_REFCOUNT_CACHE_SIZE
,
730 .type
= QEMU_OPT_SIZE
,
731 .help
= "Maximum refcount block cache size",
734 .name
= QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
735 .type
= QEMU_OPT_NUMBER
,
736 .help
= "Clean unused cache entries after this time (in seconds)",
738 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
739 "ID of secret providing qcow2 AES key or LUKS passphrase"),
740 { /* end of list */ }
744 static const char *overlap_bool_option_names
[QCOW2_OL_MAX_BITNR
] = {
745 [QCOW2_OL_MAIN_HEADER_BITNR
] = QCOW2_OPT_OVERLAP_MAIN_HEADER
,
746 [QCOW2_OL_ACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L1
,
747 [QCOW2_OL_ACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_ACTIVE_L2
,
748 [QCOW2_OL_REFCOUNT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
,
749 [QCOW2_OL_REFCOUNT_BLOCK_BITNR
] = QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
,
750 [QCOW2_OL_SNAPSHOT_TABLE_BITNR
] = QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
,
751 [QCOW2_OL_INACTIVE_L1_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L1
,
752 [QCOW2_OL_INACTIVE_L2_BITNR
] = QCOW2_OPT_OVERLAP_INACTIVE_L2
,
753 [QCOW2_OL_BITMAP_DIRECTORY_BITNR
] = QCOW2_OPT_OVERLAP_BITMAP_DIRECTORY
,
756 static void cache_clean_timer_cb(void *opaque
)
758 BlockDriverState
*bs
= opaque
;
759 BDRVQcow2State
*s
= bs
->opaque
;
760 qcow2_cache_clean_unused(s
->l2_table_cache
);
761 qcow2_cache_clean_unused(s
->refcount_block_cache
);
762 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
763 (int64_t) s
->cache_clean_interval
* 1000);
766 static void cache_clean_timer_init(BlockDriverState
*bs
, AioContext
*context
)
768 BDRVQcow2State
*s
= bs
->opaque
;
769 if (s
->cache_clean_interval
> 0) {
770 s
->cache_clean_timer
= aio_timer_new(context
, QEMU_CLOCK_VIRTUAL
,
771 SCALE_MS
, cache_clean_timer_cb
,
773 timer_mod(s
->cache_clean_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
774 (int64_t) s
->cache_clean_interval
* 1000);
778 static void cache_clean_timer_del(BlockDriverState
*bs
)
780 BDRVQcow2State
*s
= bs
->opaque
;
781 if (s
->cache_clean_timer
) {
782 timer_del(s
->cache_clean_timer
);
783 timer_free(s
->cache_clean_timer
);
784 s
->cache_clean_timer
= NULL
;
788 static void qcow2_detach_aio_context(BlockDriverState
*bs
)
790 cache_clean_timer_del(bs
);
793 static void qcow2_attach_aio_context(BlockDriverState
*bs
,
794 AioContext
*new_context
)
796 cache_clean_timer_init(bs
, new_context
);
799 static void read_cache_sizes(BlockDriverState
*bs
, QemuOpts
*opts
,
800 uint64_t *l2_cache_size
,
801 uint64_t *l2_cache_entry_size
,
802 uint64_t *refcount_cache_size
, Error
**errp
)
804 BDRVQcow2State
*s
= bs
->opaque
;
805 uint64_t combined_cache_size
, l2_cache_max_setting
;
806 bool l2_cache_size_set
, refcount_cache_size_set
, combined_cache_size_set
;
807 bool l2_cache_entry_size_set
;
808 int min_refcount_cache
= MIN_REFCOUNT_CACHE_SIZE
* s
->cluster_size
;
809 uint64_t virtual_disk_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
810 uint64_t max_l2_cache
= virtual_disk_size
/ (s
->cluster_size
/ 8);
812 combined_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_CACHE_SIZE
);
813 l2_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_SIZE
);
814 refcount_cache_size_set
= qemu_opt_get(opts
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
815 l2_cache_entry_size_set
= qemu_opt_get(opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
);
817 combined_cache_size
= qemu_opt_get_size(opts
, QCOW2_OPT_CACHE_SIZE
, 0);
818 l2_cache_max_setting
= qemu_opt_get_size(opts
, QCOW2_OPT_L2_CACHE_SIZE
,
819 DEFAULT_L2_CACHE_MAX_SIZE
);
820 *refcount_cache_size
= qemu_opt_get_size(opts
,
821 QCOW2_OPT_REFCOUNT_CACHE_SIZE
, 0);
823 *l2_cache_entry_size
= qemu_opt_get_size(
824 opts
, QCOW2_OPT_L2_CACHE_ENTRY_SIZE
, s
->cluster_size
);
826 *l2_cache_size
= MIN(max_l2_cache
, l2_cache_max_setting
);
828 if (combined_cache_size_set
) {
829 if (l2_cache_size_set
&& refcount_cache_size_set
) {
830 error_setg(errp
, QCOW2_OPT_CACHE_SIZE
", " QCOW2_OPT_L2_CACHE_SIZE
831 " and " QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not be set "
834 } else if (l2_cache_size_set
&&
835 (l2_cache_max_setting
> combined_cache_size
)) {
836 error_setg(errp
, QCOW2_OPT_L2_CACHE_SIZE
" may not exceed "
837 QCOW2_OPT_CACHE_SIZE
);
839 } else if (*refcount_cache_size
> combined_cache_size
) {
840 error_setg(errp
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
" may not exceed "
841 QCOW2_OPT_CACHE_SIZE
);
845 if (l2_cache_size_set
) {
846 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
847 } else if (refcount_cache_size_set
) {
848 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
850 /* Assign as much memory as possible to the L2 cache, and
851 * use the remainder for the refcount cache */
852 if (combined_cache_size
>= max_l2_cache
+ min_refcount_cache
) {
853 *l2_cache_size
= max_l2_cache
;
854 *refcount_cache_size
= combined_cache_size
- *l2_cache_size
;
856 *refcount_cache_size
=
857 MIN(combined_cache_size
, min_refcount_cache
);
858 *l2_cache_size
= combined_cache_size
- *refcount_cache_size
;
864 * If the L2 cache is not enough to cover the whole disk then
865 * default to 4KB entries. Smaller entries reduce the cost of
866 * loads and evictions and increase I/O performance.
868 if (*l2_cache_size
< max_l2_cache
&& !l2_cache_entry_size_set
) {
869 *l2_cache_entry_size
= MIN(s
->cluster_size
, 4096);
872 /* l2_cache_size and refcount_cache_size are ensured to have at least
873 * their minimum values in qcow2_update_options_prepare() */
875 if (*l2_cache_entry_size
< (1 << MIN_CLUSTER_BITS
) ||
876 *l2_cache_entry_size
> s
->cluster_size
||
877 !is_power_of_2(*l2_cache_entry_size
)) {
878 error_setg(errp
, "L2 cache entry size must be a power of two "
879 "between %d and the cluster size (%d)",
880 1 << MIN_CLUSTER_BITS
, s
->cluster_size
);
885 typedef struct Qcow2ReopenState
{
886 Qcow2Cache
*l2_table_cache
;
887 Qcow2Cache
*refcount_block_cache
;
888 int l2_slice_size
; /* Number of entries in a slice of the L2 table */
889 bool use_lazy_refcounts
;
891 bool discard_passthrough
[QCOW2_DISCARD_MAX
];
892 uint64_t cache_clean_interval
;
893 QCryptoBlockOpenOptions
*crypto_opts
; /* Disk encryption runtime options */
896 static int qcow2_update_options_prepare(BlockDriverState
*bs
,
898 QDict
*options
, int flags
,
901 BDRVQcow2State
*s
= bs
->opaque
;
902 QemuOpts
*opts
= NULL
;
903 const char *opt_overlap_check
, *opt_overlap_check_template
;
904 int overlap_check_template
= 0;
905 uint64_t l2_cache_size
, l2_cache_entry_size
, refcount_cache_size
;
907 const char *encryptfmt
;
908 QDict
*encryptopts
= NULL
;
909 Error
*local_err
= NULL
;
912 qdict_extract_subqdict(options
, &encryptopts
, "encrypt.");
913 encryptfmt
= qdict_get_try_str(encryptopts
, "format");
915 opts
= qemu_opts_create(&qcow2_runtime_opts
, NULL
, 0, &error_abort
);
916 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
918 error_propagate(errp
, local_err
);
923 /* get L2 table/refcount block cache size from command line options */
924 read_cache_sizes(bs
, opts
, &l2_cache_size
, &l2_cache_entry_size
,
925 &refcount_cache_size
, &local_err
);
927 error_propagate(errp
, local_err
);
932 l2_cache_size
/= l2_cache_entry_size
;
933 if (l2_cache_size
< MIN_L2_CACHE_SIZE
) {
934 l2_cache_size
= MIN_L2_CACHE_SIZE
;
936 if (l2_cache_size
> INT_MAX
) {
937 error_setg(errp
, "L2 cache size too big");
942 refcount_cache_size
/= s
->cluster_size
;
943 if (refcount_cache_size
< MIN_REFCOUNT_CACHE_SIZE
) {
944 refcount_cache_size
= MIN_REFCOUNT_CACHE_SIZE
;
946 if (refcount_cache_size
> INT_MAX
) {
947 error_setg(errp
, "Refcount cache size too big");
952 /* alloc new L2 table/refcount block cache, flush old one */
953 if (s
->l2_table_cache
) {
954 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
956 error_setg_errno(errp
, -ret
, "Failed to flush the L2 table cache");
961 if (s
->refcount_block_cache
) {
962 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
964 error_setg_errno(errp
, -ret
,
965 "Failed to flush the refcount block cache");
970 r
->l2_slice_size
= l2_cache_entry_size
/ sizeof(uint64_t);
971 r
->l2_table_cache
= qcow2_cache_create(bs
, l2_cache_size
,
972 l2_cache_entry_size
);
973 r
->refcount_block_cache
= qcow2_cache_create(bs
, refcount_cache_size
,
975 if (r
->l2_table_cache
== NULL
|| r
->refcount_block_cache
== NULL
) {
976 error_setg(errp
, "Could not allocate metadata caches");
981 /* New interval for cache cleanup timer */
982 r
->cache_clean_interval
=
983 qemu_opt_get_number(opts
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
,
984 DEFAULT_CACHE_CLEAN_INTERVAL
);
986 if (r
->cache_clean_interval
!= 0) {
987 error_setg(errp
, QCOW2_OPT_CACHE_CLEAN_INTERVAL
988 " not supported on this host");
993 if (r
->cache_clean_interval
> UINT_MAX
) {
994 error_setg(errp
, "Cache clean interval too big");
999 /* lazy-refcounts; flush if going from enabled to disabled */
1000 r
->use_lazy_refcounts
= qemu_opt_get_bool(opts
, QCOW2_OPT_LAZY_REFCOUNTS
,
1001 (s
->compatible_features
& QCOW2_COMPAT_LAZY_REFCOUNTS
));
1002 if (r
->use_lazy_refcounts
&& s
->qcow_version
< 3) {
1003 error_setg(errp
, "Lazy refcounts require a qcow2 image with at least "
1004 "qemu 1.1 compatibility level");
1009 if (s
->use_lazy_refcounts
&& !r
->use_lazy_refcounts
) {
1010 ret
= qcow2_mark_clean(bs
);
1012 error_setg_errno(errp
, -ret
, "Failed to disable lazy refcounts");
1017 /* Overlap check options */
1018 opt_overlap_check
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP
);
1019 opt_overlap_check_template
= qemu_opt_get(opts
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1020 if (opt_overlap_check_template
&& opt_overlap_check
&&
1021 strcmp(opt_overlap_check_template
, opt_overlap_check
))
1023 error_setg(errp
, "Conflicting values for qcow2 options '"
1024 QCOW2_OPT_OVERLAP
"' ('%s') and '" QCOW2_OPT_OVERLAP_TEMPLATE
1025 "' ('%s')", opt_overlap_check
, opt_overlap_check_template
);
1029 if (!opt_overlap_check
) {
1030 opt_overlap_check
= opt_overlap_check_template
?: "cached";
1033 if (!strcmp(opt_overlap_check
, "none")) {
1034 overlap_check_template
= 0;
1035 } else if (!strcmp(opt_overlap_check
, "constant")) {
1036 overlap_check_template
= QCOW2_OL_CONSTANT
;
1037 } else if (!strcmp(opt_overlap_check
, "cached")) {
1038 overlap_check_template
= QCOW2_OL_CACHED
;
1039 } else if (!strcmp(opt_overlap_check
, "all")) {
1040 overlap_check_template
= QCOW2_OL_ALL
;
1042 error_setg(errp
, "Unsupported value '%s' for qcow2 option "
1043 "'overlap-check'. Allowed are any of the following: "
1044 "none, constant, cached, all", opt_overlap_check
);
1049 r
->overlap_check
= 0;
1050 for (i
= 0; i
< QCOW2_OL_MAX_BITNR
; i
++) {
1051 /* overlap-check defines a template bitmask, but every flag may be
1052 * overwritten through the associated boolean option */
1054 qemu_opt_get_bool(opts
, overlap_bool_option_names
[i
],
1055 overlap_check_template
& (1 << i
)) << i
;
1058 r
->discard_passthrough
[QCOW2_DISCARD_NEVER
] = false;
1059 r
->discard_passthrough
[QCOW2_DISCARD_ALWAYS
] = true;
1060 r
->discard_passthrough
[QCOW2_DISCARD_REQUEST
] =
1061 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_REQUEST
,
1062 flags
& BDRV_O_UNMAP
);
1063 r
->discard_passthrough
[QCOW2_DISCARD_SNAPSHOT
] =
1064 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_SNAPSHOT
, true);
1065 r
->discard_passthrough
[QCOW2_DISCARD_OTHER
] =
1066 qemu_opt_get_bool(opts
, QCOW2_OPT_DISCARD_OTHER
, false);
1068 switch (s
->crypt_method_header
) {
1069 case QCOW_CRYPT_NONE
:
1071 error_setg(errp
, "No encryption in image header, but options "
1072 "specified format '%s'", encryptfmt
);
1078 case QCOW_CRYPT_AES
:
1079 if (encryptfmt
&& !g_str_equal(encryptfmt
, "aes")) {
1081 "Header reported 'aes' encryption format but "
1082 "options specify '%s'", encryptfmt
);
1086 qdict_put_str(encryptopts
, "format", "qcow");
1087 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1090 case QCOW_CRYPT_LUKS
:
1091 if (encryptfmt
&& !g_str_equal(encryptfmt
, "luks")) {
1093 "Header reported 'luks' encryption format but "
1094 "options specify '%s'", encryptfmt
);
1098 qdict_put_str(encryptopts
, "format", "luks");
1099 r
->crypto_opts
= block_crypto_open_opts_init(encryptopts
, errp
);
1103 error_setg(errp
, "Unsupported encryption method %d",
1104 s
->crypt_method_header
);
1107 if (s
->crypt_method_header
!= QCOW_CRYPT_NONE
&& !r
->crypto_opts
) {
1114 qobject_unref(encryptopts
);
1115 qemu_opts_del(opts
);
1120 static void qcow2_update_options_commit(BlockDriverState
*bs
,
1121 Qcow2ReopenState
*r
)
1123 BDRVQcow2State
*s
= bs
->opaque
;
1126 if (s
->l2_table_cache
) {
1127 qcow2_cache_destroy(s
->l2_table_cache
);
1129 if (s
->refcount_block_cache
) {
1130 qcow2_cache_destroy(s
->refcount_block_cache
);
1132 s
->l2_table_cache
= r
->l2_table_cache
;
1133 s
->refcount_block_cache
= r
->refcount_block_cache
;
1134 s
->l2_slice_size
= r
->l2_slice_size
;
1136 s
->overlap_check
= r
->overlap_check
;
1137 s
->use_lazy_refcounts
= r
->use_lazy_refcounts
;
1139 for (i
= 0; i
< QCOW2_DISCARD_MAX
; i
++) {
1140 s
->discard_passthrough
[i
] = r
->discard_passthrough
[i
];
1143 if (s
->cache_clean_interval
!= r
->cache_clean_interval
) {
1144 cache_clean_timer_del(bs
);
1145 s
->cache_clean_interval
= r
->cache_clean_interval
;
1146 cache_clean_timer_init(bs
, bdrv_get_aio_context(bs
));
1149 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1150 s
->crypto_opts
= r
->crypto_opts
;
1153 static void qcow2_update_options_abort(BlockDriverState
*bs
,
1154 Qcow2ReopenState
*r
)
1156 if (r
->l2_table_cache
) {
1157 qcow2_cache_destroy(r
->l2_table_cache
);
1159 if (r
->refcount_block_cache
) {
1160 qcow2_cache_destroy(r
->refcount_block_cache
);
1162 qapi_free_QCryptoBlockOpenOptions(r
->crypto_opts
);
1165 static int qcow2_update_options(BlockDriverState
*bs
, QDict
*options
,
1166 int flags
, Error
**errp
)
1168 Qcow2ReopenState r
= {};
1171 ret
= qcow2_update_options_prepare(bs
, &r
, options
, flags
, errp
);
1173 qcow2_update_options_commit(bs
, &r
);
1175 qcow2_update_options_abort(bs
, &r
);
1181 /* Called with s->lock held. */
1182 static int coroutine_fn
qcow2_do_open(BlockDriverState
*bs
, QDict
*options
,
1183 int flags
, Error
**errp
)
1185 BDRVQcow2State
*s
= bs
->opaque
;
1186 unsigned int len
, i
;
1189 Error
*local_err
= NULL
;
1191 uint64_t l1_vm_state_index
;
1192 bool update_header
= false;
1194 ret
= bdrv_pread(bs
->file
, 0, &header
, sizeof(header
));
1196 error_setg_errno(errp
, -ret
, "Could not read qcow2 header");
1199 header
.magic
= be32_to_cpu(header
.magic
);
1200 header
.version
= be32_to_cpu(header
.version
);
1201 header
.backing_file_offset
= be64_to_cpu(header
.backing_file_offset
);
1202 header
.backing_file_size
= be32_to_cpu(header
.backing_file_size
);
1203 header
.size
= be64_to_cpu(header
.size
);
1204 header
.cluster_bits
= be32_to_cpu(header
.cluster_bits
);
1205 header
.crypt_method
= be32_to_cpu(header
.crypt_method
);
1206 header
.l1_table_offset
= be64_to_cpu(header
.l1_table_offset
);
1207 header
.l1_size
= be32_to_cpu(header
.l1_size
);
1208 header
.refcount_table_offset
= be64_to_cpu(header
.refcount_table_offset
);
1209 header
.refcount_table_clusters
=
1210 be32_to_cpu(header
.refcount_table_clusters
);
1211 header
.snapshots_offset
= be64_to_cpu(header
.snapshots_offset
);
1212 header
.nb_snapshots
= be32_to_cpu(header
.nb_snapshots
);
1214 if (header
.magic
!= QCOW_MAGIC
) {
1215 error_setg(errp
, "Image is not in qcow2 format");
1219 if (header
.version
< 2 || header
.version
> 3) {
1220 error_setg(errp
, "Unsupported qcow2 version %" PRIu32
, header
.version
);
1225 s
->qcow_version
= header
.version
;
1227 /* Initialise cluster size */
1228 if (header
.cluster_bits
< MIN_CLUSTER_BITS
||
1229 header
.cluster_bits
> MAX_CLUSTER_BITS
) {
1230 error_setg(errp
, "Unsupported cluster size: 2^%" PRIu32
,
1231 header
.cluster_bits
);
1236 s
->cluster_bits
= header
.cluster_bits
;
1237 s
->cluster_size
= 1 << s
->cluster_bits
;
1238 s
->cluster_sectors
= 1 << (s
->cluster_bits
- BDRV_SECTOR_BITS
);
1240 /* Initialise version 3 header fields */
1241 if (header
.version
== 2) {
1242 header
.incompatible_features
= 0;
1243 header
.compatible_features
= 0;
1244 header
.autoclear_features
= 0;
1245 header
.refcount_order
= 4;
1246 header
.header_length
= 72;
1248 header
.incompatible_features
=
1249 be64_to_cpu(header
.incompatible_features
);
1250 header
.compatible_features
= be64_to_cpu(header
.compatible_features
);
1251 header
.autoclear_features
= be64_to_cpu(header
.autoclear_features
);
1252 header
.refcount_order
= be32_to_cpu(header
.refcount_order
);
1253 header
.header_length
= be32_to_cpu(header
.header_length
);
1255 if (header
.header_length
< 104) {
1256 error_setg(errp
, "qcow2 header too short");
1262 if (header
.header_length
> s
->cluster_size
) {
1263 error_setg(errp
, "qcow2 header exceeds cluster size");
1268 if (header
.header_length
> sizeof(header
)) {
1269 s
->unknown_header_fields_size
= header
.header_length
- sizeof(header
);
1270 s
->unknown_header_fields
= g_malloc(s
->unknown_header_fields_size
);
1271 ret
= bdrv_pread(bs
->file
, sizeof(header
), s
->unknown_header_fields
,
1272 s
->unknown_header_fields_size
);
1274 error_setg_errno(errp
, -ret
, "Could not read unknown qcow2 header "
1280 if (header
.backing_file_offset
> s
->cluster_size
) {
1281 error_setg(errp
, "Invalid backing file offset");
1286 if (header
.backing_file_offset
) {
1287 ext_end
= header
.backing_file_offset
;
1289 ext_end
= 1 << header
.cluster_bits
;
1292 /* Handle feature bits */
1293 s
->incompatible_features
= header
.incompatible_features
;
1294 s
->compatible_features
= header
.compatible_features
;
1295 s
->autoclear_features
= header
.autoclear_features
;
1297 if (s
->incompatible_features
& ~QCOW2_INCOMPAT_MASK
) {
1298 void *feature_table
= NULL
;
1299 qcow2_read_extensions(bs
, header
.header_length
, ext_end
,
1300 &feature_table
, flags
, NULL
, NULL
);
1301 report_unsupported_feature(errp
, feature_table
,
1302 s
->incompatible_features
&
1303 ~QCOW2_INCOMPAT_MASK
);
1305 g_free(feature_table
);
1309 if (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
) {
1310 /* Corrupt images may not be written to unless they are being repaired
1312 if ((flags
& BDRV_O_RDWR
) && !(flags
& BDRV_O_CHECK
)) {
1313 error_setg(errp
, "qcow2: Image is corrupt; cannot be opened "
1320 /* Check support for various header values */
1321 if (header
.refcount_order
> 6) {
1322 error_setg(errp
, "Reference count entry width too large; may not "
1327 s
->refcount_order
= header
.refcount_order
;
1328 s
->refcount_bits
= 1 << s
->refcount_order
;
1329 s
->refcount_max
= UINT64_C(1) << (s
->refcount_bits
- 1);
1330 s
->refcount_max
+= s
->refcount_max
- 1;
1332 s
->crypt_method_header
= header
.crypt_method
;
1333 if (s
->crypt_method_header
) {
1334 if (bdrv_uses_whitelist() &&
1335 s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1337 "Use of AES-CBC encrypted qcow2 images is no longer "
1338 "supported in system emulators");
1339 error_append_hint(errp
,
1340 "You can use 'qemu-img convert' to convert your "
1341 "image to an alternative supported format, such "
1342 "as unencrypted qcow2, or raw with the LUKS "
1343 "format instead.\n");
1348 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1349 s
->crypt_physical_offset
= false;
1351 /* Assuming LUKS and any future crypt methods we
1352 * add will all use physical offsets, due to the
1353 * fact that the alternative is insecure... */
1354 s
->crypt_physical_offset
= true;
1357 bs
->encrypted
= true;
1360 s
->l2_bits
= s
->cluster_bits
- 3; /* L2 is always one cluster */
1361 s
->l2_size
= 1 << s
->l2_bits
;
1362 /* 2^(s->refcount_order - 3) is the refcount width in bytes */
1363 s
->refcount_block_bits
= s
->cluster_bits
- (s
->refcount_order
- 3);
1364 s
->refcount_block_size
= 1 << s
->refcount_block_bits
;
1365 bs
->total_sectors
= header
.size
/ BDRV_SECTOR_SIZE
;
1366 s
->csize_shift
= (62 - (s
->cluster_bits
- 8));
1367 s
->csize_mask
= (1 << (s
->cluster_bits
- 8)) - 1;
1368 s
->cluster_offset_mask
= (1LL << s
->csize_shift
) - 1;
1370 s
->refcount_table_offset
= header
.refcount_table_offset
;
1371 s
->refcount_table_size
=
1372 header
.refcount_table_clusters
<< (s
->cluster_bits
- 3);
1374 if (header
.refcount_table_clusters
== 0 && !(flags
& BDRV_O_CHECK
)) {
1375 error_setg(errp
, "Image does not contain a reference count table");
1380 ret
= qcow2_validate_table(bs
, s
->refcount_table_offset
,
1381 header
.refcount_table_clusters
,
1382 s
->cluster_size
, QCOW_MAX_REFTABLE_SIZE
,
1383 "Reference count table", errp
);
1388 /* The total size in bytes of the snapshot table is checked in
1389 * qcow2_read_snapshots() because the size of each snapshot is
1390 * variable and we don't know it yet.
1391 * Here we only check the offset and number of snapshots. */
1392 ret
= qcow2_validate_table(bs
, header
.snapshots_offset
,
1393 header
.nb_snapshots
,
1394 sizeof(QCowSnapshotHeader
),
1395 sizeof(QCowSnapshotHeader
) * QCOW_MAX_SNAPSHOTS
,
1396 "Snapshot table", errp
);
1401 /* read the level 1 table */
1402 ret
= qcow2_validate_table(bs
, header
.l1_table_offset
,
1403 header
.l1_size
, sizeof(uint64_t),
1404 QCOW_MAX_L1_SIZE
, "Active L1 table", errp
);
1408 s
->l1_size
= header
.l1_size
;
1409 s
->l1_table_offset
= header
.l1_table_offset
;
1411 l1_vm_state_index
= size_to_l1(s
, header
.size
);
1412 if (l1_vm_state_index
> INT_MAX
) {
1413 error_setg(errp
, "Image is too big");
1417 s
->l1_vm_state_index
= l1_vm_state_index
;
1419 /* the L1 table must contain at least enough entries to put
1420 header.size bytes */
1421 if (s
->l1_size
< s
->l1_vm_state_index
) {
1422 error_setg(errp
, "L1 table is too small");
1427 if (s
->l1_size
> 0) {
1428 s
->l1_table
= qemu_try_blockalign(bs
->file
->bs
,
1429 ROUND_UP(s
->l1_size
* sizeof(uint64_t), 512));
1430 if (s
->l1_table
== NULL
) {
1431 error_setg(errp
, "Could not allocate L1 table");
1435 ret
= bdrv_pread(bs
->file
, s
->l1_table_offset
, s
->l1_table
,
1436 s
->l1_size
* sizeof(uint64_t));
1438 error_setg_errno(errp
, -ret
, "Could not read L1 table");
1441 for(i
= 0;i
< s
->l1_size
; i
++) {
1442 s
->l1_table
[i
] = be64_to_cpu(s
->l1_table
[i
]);
1446 /* Parse driver-specific options */
1447 ret
= qcow2_update_options(bs
, options
, flags
, errp
);
1454 ret
= qcow2_refcount_init(bs
);
1456 error_setg_errno(errp
, -ret
, "Could not initialize refcount handling");
1460 QLIST_INIT(&s
->cluster_allocs
);
1461 QTAILQ_INIT(&s
->discards
);
1463 /* read qcow2 extensions */
1464 if (qcow2_read_extensions(bs
, header
.header_length
, ext_end
, NULL
,
1465 flags
, &update_header
, &local_err
)) {
1466 error_propagate(errp
, local_err
);
1471 /* Open external data file */
1472 s
->data_file
= bdrv_open_child(NULL
, options
, "data-file", bs
, &child_file
,
1475 error_propagate(errp
, local_err
);
1480 if (s
->incompatible_features
& QCOW2_INCOMPAT_DATA_FILE
) {
1481 if (!s
->data_file
&& s
->image_data_file
) {
1482 s
->data_file
= bdrv_open_child(s
->image_data_file
, options
,
1483 "data-file", bs
, &child_file
,
1485 if (!s
->data_file
) {
1490 if (!s
->data_file
) {
1491 error_setg(errp
, "'data-file' is required for this image");
1497 error_setg(errp
, "'data-file' can only be set for images with an "
1498 "external data file");
1503 s
->data_file
= bs
->file
;
1505 if (data_file_is_raw(bs
)) {
1506 error_setg(errp
, "data-file-raw requires a data file");
1512 /* qcow2_read_extension may have set up the crypto context
1513 * if the crypt method needs a header region, some methods
1514 * don't need header extensions, so must check here
1516 if (s
->crypt_method_header
&& !s
->crypto
) {
1517 if (s
->crypt_method_header
== QCOW_CRYPT_AES
) {
1518 unsigned int cflags
= 0;
1519 if (flags
& BDRV_O_NO_IO
) {
1520 cflags
|= QCRYPTO_BLOCK_OPEN_NO_IO
;
1522 s
->crypto
= qcrypto_block_open(s
->crypto_opts
, "encrypt.",
1523 NULL
, NULL
, cflags
, 1, errp
);
1528 } else if (!(flags
& BDRV_O_NO_IO
)) {
1529 error_setg(errp
, "Missing CRYPTO header for crypt method %d",
1530 s
->crypt_method_header
);
1536 /* read the backing file name */
1537 if (header
.backing_file_offset
!= 0) {
1538 len
= header
.backing_file_size
;
1539 if (len
> MIN(1023, s
->cluster_size
- header
.backing_file_offset
) ||
1540 len
>= sizeof(bs
->backing_file
)) {
1541 error_setg(errp
, "Backing file name too long");
1545 ret
= bdrv_pread(bs
->file
, header
.backing_file_offset
,
1546 bs
->auto_backing_file
, len
);
1548 error_setg_errno(errp
, -ret
, "Could not read backing file name");
1551 bs
->auto_backing_file
[len
] = '\0';
1552 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
),
1553 bs
->auto_backing_file
);
1554 s
->image_backing_file
= g_strdup(bs
->auto_backing_file
);
1557 /* Internal snapshots */
1558 s
->snapshots_offset
= header
.snapshots_offset
;
1559 s
->nb_snapshots
= header
.nb_snapshots
;
1561 ret
= qcow2_read_snapshots(bs
);
1563 error_setg_errno(errp
, -ret
, "Could not read snapshots");
1567 /* Clear unknown autoclear feature bits */
1568 update_header
|= s
->autoclear_features
& ~QCOW2_AUTOCLEAR_MASK
;
1570 update_header
&& !bs
->read_only
&& !(flags
& BDRV_O_INACTIVE
);
1571 if (update_header
) {
1572 s
->autoclear_features
&= QCOW2_AUTOCLEAR_MASK
;
1575 /* == Handle persistent dirty bitmaps ==
1577 * We want load dirty bitmaps in three cases:
1579 * 1. Normal open of the disk in active mode, not related to invalidation
1582 * 2. Invalidation of the target vm after pre-copy phase of migration, if
1583 * bitmaps are _not_ migrating through migration channel, i.e.
1584 * 'dirty-bitmaps' capability is disabled.
1586 * 3. Invalidation of source vm after failed or canceled migration.
1587 * This is a very interesting case. There are two possible types of
1590 * A. Stored on inactivation and removed. They should be loaded from the
1593 * B. Not stored: not-persistent bitmaps and bitmaps, migrated through
1594 * the migration channel (with dirty-bitmaps capability).
1596 * On the other hand, there are two possible sub-cases:
1598 * 3.1 disk was changed by somebody else while were inactive. In this
1599 * case all in-RAM dirty bitmaps (both persistent and not) are
1600 * definitely invalid. And we don't have any method to determine
1603 * Simple and safe thing is to just drop all the bitmaps of type B on
1604 * inactivation. But in this case we lose bitmaps in valid 4.2 case.
1606 * On the other hand, resuming source vm, if disk was already changed
1607 * is a bad thing anyway: not only bitmaps, the whole vm state is
1608 * out of sync with disk.
1610 * This means, that user or management tool, who for some reason
1611 * decided to resume source vm, after disk was already changed by
1612 * target vm, should at least drop all dirty bitmaps by hand.
1614 * So, we can ignore this case for now, but TODO: "generation"
1615 * extension for qcow2, to determine, that image was changed after
1616 * last inactivation. And if it is changed, we will drop (or at least
1617 * mark as 'invalid' all the bitmaps of type B, both persistent
1620 * 3.2 disk was _not_ changed while were inactive. Bitmaps may be saved
1621 * to disk ('dirty-bitmaps' capability disabled), or not saved
1622 * ('dirty-bitmaps' capability enabled), but we don't need to care
1623 * of: let's load bitmaps as always: stored bitmaps will be loaded,
1624 * and not stored has flag IN_USE=1 in the image and will be skipped
1627 * One remaining possible case when we don't want load bitmaps:
1629 * 4. Open disk in inactive mode in target vm (bitmaps are migrating or
1630 * will be loaded on invalidation, no needs try loading them before)
1633 if (!(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
)) {
1634 /* It's case 1, 2 or 3.2. Or 3.1 which is BUG in management layer. */
1635 bool header_updated
= qcow2_load_dirty_bitmaps(bs
, &local_err
);
1637 update_header
= update_header
&& !header_updated
;
1639 if (local_err
!= NULL
) {
1640 error_propagate(errp
, local_err
);
1645 if (update_header
) {
1646 ret
= qcow2_update_header(bs
);
1648 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
1653 bs
->supported_zero_flags
= header
.version
>= 3 ? BDRV_REQ_MAY_UNMAP
: 0;
1655 /* Repair image if dirty */
1656 if (!(flags
& (BDRV_O_CHECK
| BDRV_O_INACTIVE
)) && !bs
->read_only
&&
1657 (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
)) {
1658 BdrvCheckResult result
= {0};
1660 ret
= qcow2_co_check_locked(bs
, &result
,
1661 BDRV_FIX_ERRORS
| BDRV_FIX_LEAKS
);
1662 if (ret
< 0 || result
.check_errors
) {
1666 error_setg_errno(errp
, -ret
, "Could not repair dirty image");
1673 BdrvCheckResult result
= {0};
1674 qcow2_check_refcounts(bs
, &result
, 0);
1678 qemu_co_queue_init(&s
->compress_wait_queue
);
1683 g_free(s
->image_data_file
);
1684 if (has_data_file(bs
)) {
1685 bdrv_unref_child(bs
, s
->data_file
);
1687 g_free(s
->unknown_header_fields
);
1688 cleanup_unknown_header_ext(bs
);
1689 qcow2_free_snapshots(bs
);
1690 qcow2_refcount_close(bs
);
1691 qemu_vfree(s
->l1_table
);
1692 /* else pre-write overlap checks in cache_destroy may crash */
1694 cache_clean_timer_del(bs
);
1695 if (s
->l2_table_cache
) {
1696 qcow2_cache_destroy(s
->l2_table_cache
);
1698 if (s
->refcount_block_cache
) {
1699 qcow2_cache_destroy(s
->refcount_block_cache
);
1701 qcrypto_block_free(s
->crypto
);
1702 qapi_free_QCryptoBlockOpenOptions(s
->crypto_opts
);
1706 typedef struct QCow2OpenCo
{
1707 BlockDriverState
*bs
;
1714 static void coroutine_fn
qcow2_open_entry(void *opaque
)
1716 QCow2OpenCo
*qoc
= opaque
;
1717 BDRVQcow2State
*s
= qoc
->bs
->opaque
;
1719 qemu_co_mutex_lock(&s
->lock
);
1720 qoc
->ret
= qcow2_do_open(qoc
->bs
, qoc
->options
, qoc
->flags
, qoc
->errp
);
1721 qemu_co_mutex_unlock(&s
->lock
);
1724 static int qcow2_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1727 BDRVQcow2State
*s
= bs
->opaque
;
1736 bs
->file
= bdrv_open_child(NULL
, options
, "file", bs
, &child_file
,
1742 /* Initialise locks */
1743 qemu_co_mutex_init(&s
->lock
);
1745 if (qemu_in_coroutine()) {
1746 /* From bdrv_co_create. */
1747 qcow2_open_entry(&qoc
);
1749 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1750 qemu_coroutine_enter(qemu_coroutine_create(qcow2_open_entry
, &qoc
));
1751 BDRV_POLL_WHILE(bs
, qoc
.ret
== -EINPROGRESS
);
1756 static void qcow2_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1758 BDRVQcow2State
*s
= bs
->opaque
;
1760 if (bs
->encrypted
) {
1761 /* Encryption works on a sector granularity */
1762 bs
->bl
.request_alignment
= qcrypto_block_get_sector_size(s
->crypto
);
1764 bs
->bl
.pwrite_zeroes_alignment
= s
->cluster_size
;
1765 bs
->bl
.pdiscard_alignment
= s
->cluster_size
;
1768 static int qcow2_reopen_prepare(BDRVReopenState
*state
,
1769 BlockReopenQueue
*queue
, Error
**errp
)
1771 Qcow2ReopenState
*r
;
1774 r
= g_new0(Qcow2ReopenState
, 1);
1777 ret
= qcow2_update_options_prepare(state
->bs
, r
, state
->options
,
1778 state
->flags
, errp
);
1783 /* We need to write out any unwritten data if we reopen read-only. */
1784 if ((state
->flags
& BDRV_O_RDWR
) == 0) {
1785 ret
= qcow2_reopen_bitmaps_ro(state
->bs
, errp
);
1790 ret
= bdrv_flush(state
->bs
);
1795 ret
= qcow2_mark_clean(state
->bs
);
1804 qcow2_update_options_abort(state
->bs
, r
);
1809 static void qcow2_reopen_commit(BDRVReopenState
*state
)
1811 qcow2_update_options_commit(state
->bs
, state
->opaque
);
1812 g_free(state
->opaque
);
1815 static void qcow2_reopen_abort(BDRVReopenState
*state
)
1817 qcow2_update_options_abort(state
->bs
, state
->opaque
);
1818 g_free(state
->opaque
);
1821 static void qcow2_join_options(QDict
*options
, QDict
*old_options
)
1823 bool has_new_overlap_template
=
1824 qdict_haskey(options
, QCOW2_OPT_OVERLAP
) ||
1825 qdict_haskey(options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1826 bool has_new_total_cache_size
=
1827 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
);
1828 bool has_all_cache_options
;
1830 /* New overlap template overrides all old overlap options */
1831 if (has_new_overlap_template
) {
1832 qdict_del(old_options
, QCOW2_OPT_OVERLAP
);
1833 qdict_del(old_options
, QCOW2_OPT_OVERLAP_TEMPLATE
);
1834 qdict_del(old_options
, QCOW2_OPT_OVERLAP_MAIN_HEADER
);
1835 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L1
);
1836 qdict_del(old_options
, QCOW2_OPT_OVERLAP_ACTIVE_L2
);
1837 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_TABLE
);
1838 qdict_del(old_options
, QCOW2_OPT_OVERLAP_REFCOUNT_BLOCK
);
1839 qdict_del(old_options
, QCOW2_OPT_OVERLAP_SNAPSHOT_TABLE
);
1840 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L1
);
1841 qdict_del(old_options
, QCOW2_OPT_OVERLAP_INACTIVE_L2
);
1844 /* New total cache size overrides all old options */
1845 if (qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
)) {
1846 qdict_del(old_options
, QCOW2_OPT_L2_CACHE_SIZE
);
1847 qdict_del(old_options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1850 qdict_join(options
, old_options
, false);
1853 * If after merging all cache size options are set, an old total size is
1854 * overwritten. Do keep all options, however, if all three are new. The
1855 * resulting error message is what we want to happen.
1857 has_all_cache_options
=
1858 qdict_haskey(options
, QCOW2_OPT_CACHE_SIZE
) ||
1859 qdict_haskey(options
, QCOW2_OPT_L2_CACHE_SIZE
) ||
1860 qdict_haskey(options
, QCOW2_OPT_REFCOUNT_CACHE_SIZE
);
1862 if (has_all_cache_options
&& !has_new_total_cache_size
) {
1863 qdict_del(options
, QCOW2_OPT_CACHE_SIZE
);
1867 static int coroutine_fn
qcow2_co_block_status(BlockDriverState
*bs
,
1869 int64_t offset
, int64_t count
,
1870 int64_t *pnum
, int64_t *map
,
1871 BlockDriverState
**file
)
1873 BDRVQcow2State
*s
= bs
->opaque
;
1874 uint64_t cluster_offset
;
1875 int index_in_cluster
, ret
;
1879 bytes
= MIN(INT_MAX
, count
);
1880 qemu_co_mutex_lock(&s
->lock
);
1881 ret
= qcow2_get_cluster_offset(bs
, offset
, &bytes
, &cluster_offset
);
1882 qemu_co_mutex_unlock(&s
->lock
);
1889 if ((ret
== QCOW2_CLUSTER_NORMAL
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) &&
1891 index_in_cluster
= offset
& (s
->cluster_size
- 1);
1892 *map
= cluster_offset
| index_in_cluster
;
1893 *file
= s
->data_file
->bs
;
1894 status
|= BDRV_BLOCK_OFFSET_VALID
;
1896 if (ret
== QCOW2_CLUSTER_ZERO_PLAIN
|| ret
== QCOW2_CLUSTER_ZERO_ALLOC
) {
1897 status
|= BDRV_BLOCK_ZERO
;
1898 } else if (ret
!= QCOW2_CLUSTER_UNALLOCATED
) {
1899 status
|= BDRV_BLOCK_DATA
;
1904 static coroutine_fn
int qcow2_handle_l2meta(BlockDriverState
*bs
,
1905 QCowL2Meta
**pl2meta
,
1909 QCowL2Meta
*l2meta
= *pl2meta
;
1911 while (l2meta
!= NULL
) {
1915 ret
= qcow2_alloc_cluster_link_l2(bs
, l2meta
);
1920 qcow2_alloc_cluster_abort(bs
, l2meta
);
1923 /* Take the request off the list of running requests */
1924 if (l2meta
->nb_clusters
!= 0) {
1925 QLIST_REMOVE(l2meta
, next_in_flight
);
1928 qemu_co_queue_restart_all(&l2meta
->dependent_requests
);
1930 next
= l2meta
->next
;
1939 static coroutine_fn
int qcow2_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1940 uint64_t bytes
, QEMUIOVector
*qiov
,
1943 BDRVQcow2State
*s
= bs
->opaque
;
1944 int offset_in_cluster
;
1946 unsigned int cur_bytes
; /* number of bytes in current iteration */
1947 uint64_t cluster_offset
= 0;
1948 uint64_t bytes_done
= 0;
1949 QEMUIOVector hd_qiov
;
1950 uint8_t *cluster_data
= NULL
;
1952 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
1954 qemu_co_mutex_lock(&s
->lock
);
1956 while (bytes
!= 0) {
1958 /* prepare next request */
1959 cur_bytes
= MIN(bytes
, INT_MAX
);
1961 cur_bytes
= MIN(cur_bytes
,
1962 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
1965 ret
= qcow2_get_cluster_offset(bs
, offset
, &cur_bytes
, &cluster_offset
);
1970 offset_in_cluster
= offset_into_cluster(s
, offset
);
1972 qemu_iovec_reset(&hd_qiov
);
1973 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
1976 case QCOW2_CLUSTER_UNALLOCATED
:
1979 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_BACKING_AIO
);
1980 qemu_co_mutex_unlock(&s
->lock
);
1981 ret
= bdrv_co_preadv(bs
->backing
, offset
, cur_bytes
,
1983 qemu_co_mutex_lock(&s
->lock
);
1988 /* Note: in this case, no need to wait */
1989 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1993 case QCOW2_CLUSTER_ZERO_PLAIN
:
1994 case QCOW2_CLUSTER_ZERO_ALLOC
:
1995 qemu_iovec_memset(&hd_qiov
, 0, 0, cur_bytes
);
1998 case QCOW2_CLUSTER_COMPRESSED
:
1999 qemu_co_mutex_unlock(&s
->lock
);
2000 ret
= qcow2_co_preadv_compressed(bs
, cluster_offset
,
2003 qemu_co_mutex_lock(&s
->lock
);
2010 case QCOW2_CLUSTER_NORMAL
:
2011 if ((cluster_offset
& 511) != 0) {
2016 if (bs
->encrypted
) {
2020 * For encrypted images, read everything into a temporary
2021 * contiguous buffer on which the AES functions can work.
2023 if (!cluster_data
) {
2025 qemu_try_blockalign(s
->data_file
->bs
,
2026 QCOW_MAX_CRYPT_CLUSTERS
2028 if (cluster_data
== NULL
) {
2034 assert(cur_bytes
<= QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2035 qemu_iovec_reset(&hd_qiov
);
2036 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2039 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_AIO
);
2040 qemu_co_mutex_unlock(&s
->lock
);
2041 ret
= bdrv_co_preadv(s
->data_file
,
2042 cluster_offset
+ offset_in_cluster
,
2043 cur_bytes
, &hd_qiov
, 0);
2044 qemu_co_mutex_lock(&s
->lock
);
2048 if (bs
->encrypted
) {
2050 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
2051 assert((cur_bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
2052 if (qcrypto_block_decrypt(s
->crypto
,
2053 (s
->crypt_physical_offset
?
2054 cluster_offset
+ offset_in_cluster
:
2062 qemu_iovec_from_buf(qiov
, bytes_done
, cluster_data
, cur_bytes
);
2067 g_assert_not_reached();
2073 offset
+= cur_bytes
;
2074 bytes_done
+= cur_bytes
;
2079 qemu_co_mutex_unlock(&s
->lock
);
2081 qemu_iovec_destroy(&hd_qiov
);
2082 qemu_vfree(cluster_data
);
2087 /* Check if it's possible to merge a write request with the writing of
2088 * the data from the COW regions */
2089 static bool merge_cow(uint64_t offset
, unsigned bytes
,
2090 QEMUIOVector
*hd_qiov
, QCowL2Meta
*l2meta
)
2094 for (m
= l2meta
; m
!= NULL
; m
= m
->next
) {
2095 /* If both COW regions are empty then there's nothing to merge */
2096 if (m
->cow_start
.nb_bytes
== 0 && m
->cow_end
.nb_bytes
== 0) {
2100 /* The data (middle) region must be immediately after the
2102 if (l2meta_cow_start(m
) + m
->cow_start
.nb_bytes
!= offset
) {
2106 /* The end region must be immediately after the data (middle)
2108 if (m
->offset
+ m
->cow_end
.offset
!= offset
+ bytes
) {
2112 /* Make sure that adding both COW regions to the QEMUIOVector
2113 * does not exceed IOV_MAX */
2114 if (hd_qiov
->niov
> IOV_MAX
- 2) {
2118 m
->data_qiov
= hd_qiov
;
2125 static coroutine_fn
int qcow2_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
2126 uint64_t bytes
, QEMUIOVector
*qiov
,
2129 BDRVQcow2State
*s
= bs
->opaque
;
2130 int offset_in_cluster
;
2132 unsigned int cur_bytes
; /* number of sectors in current iteration */
2133 uint64_t cluster_offset
;
2134 QEMUIOVector hd_qiov
;
2135 uint64_t bytes_done
= 0;
2136 uint8_t *cluster_data
= NULL
;
2137 QCowL2Meta
*l2meta
= NULL
;
2139 trace_qcow2_writev_start_req(qemu_coroutine_self(), offset
, bytes
);
2141 qemu_iovec_init(&hd_qiov
, qiov
->niov
);
2143 qemu_co_mutex_lock(&s
->lock
);
2145 while (bytes
!= 0) {
2149 trace_qcow2_writev_start_part(qemu_coroutine_self());
2150 offset_in_cluster
= offset_into_cluster(s
, offset
);
2151 cur_bytes
= MIN(bytes
, INT_MAX
);
2152 if (bs
->encrypted
) {
2153 cur_bytes
= MIN(cur_bytes
,
2154 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
2155 - offset_in_cluster
);
2158 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2159 &cluster_offset
, &l2meta
);
2164 assert((cluster_offset
& 511) == 0);
2166 qemu_iovec_reset(&hd_qiov
);
2167 qemu_iovec_concat(&hd_qiov
, qiov
, bytes_done
, cur_bytes
);
2169 if (bs
->encrypted
) {
2171 if (!cluster_data
) {
2172 cluster_data
= qemu_try_blockalign(bs
->file
->bs
,
2173 QCOW_MAX_CRYPT_CLUSTERS
2175 if (cluster_data
== NULL
) {
2181 assert(hd_qiov
.size
<=
2182 QCOW_MAX_CRYPT_CLUSTERS
* s
->cluster_size
);
2183 qemu_iovec_to_buf(&hd_qiov
, 0, cluster_data
, hd_qiov
.size
);
2185 if (qcrypto_block_encrypt(s
->crypto
,
2186 (s
->crypt_physical_offset
?
2187 cluster_offset
+ offset_in_cluster
:
2190 cur_bytes
, NULL
) < 0) {
2195 qemu_iovec_reset(&hd_qiov
);
2196 qemu_iovec_add(&hd_qiov
, cluster_data
, cur_bytes
);
2199 ret
= qcow2_pre_write_overlap_check(bs
, 0,
2200 cluster_offset
+ offset_in_cluster
, cur_bytes
, true);
2205 /* If we need to do COW, check if it's possible to merge the
2206 * writing of the guest data together with that of the COW regions.
2207 * If it's not possible (or not necessary) then write the
2208 * guest data now. */
2209 if (!merge_cow(offset
, cur_bytes
, &hd_qiov
, l2meta
)) {
2210 qemu_co_mutex_unlock(&s
->lock
);
2211 BLKDBG_EVENT(bs
->file
, BLKDBG_WRITE_AIO
);
2212 trace_qcow2_writev_data(qemu_coroutine_self(),
2213 cluster_offset
+ offset_in_cluster
);
2214 ret
= bdrv_co_pwritev(s
->data_file
,
2215 cluster_offset
+ offset_in_cluster
,
2216 cur_bytes
, &hd_qiov
, 0);
2217 qemu_co_mutex_lock(&s
->lock
);
2223 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
2229 offset
+= cur_bytes
;
2230 bytes_done
+= cur_bytes
;
2231 trace_qcow2_writev_done_part(qemu_coroutine_self(), cur_bytes
);
2236 qcow2_handle_l2meta(bs
, &l2meta
, false);
2238 qemu_co_mutex_unlock(&s
->lock
);
2240 qemu_iovec_destroy(&hd_qiov
);
2241 qemu_vfree(cluster_data
);
2242 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
2247 static int qcow2_inactivate(BlockDriverState
*bs
)
2249 BDRVQcow2State
*s
= bs
->opaque
;
2250 int ret
, result
= 0;
2251 Error
*local_err
= NULL
;
2253 qcow2_store_persistent_dirty_bitmaps(bs
, &local_err
);
2254 if (local_err
!= NULL
) {
2256 error_reportf_err(local_err
, "Lost persistent bitmaps during "
2257 "inactivation of node '%s': ",
2258 bdrv_get_device_or_node_name(bs
));
2261 ret
= qcow2_cache_flush(bs
, s
->l2_table_cache
);
2264 error_report("Failed to flush the L2 table cache: %s",
2268 ret
= qcow2_cache_flush(bs
, s
->refcount_block_cache
);
2271 error_report("Failed to flush the refcount block cache: %s",
2276 qcow2_mark_clean(bs
);
2282 static void qcow2_close(BlockDriverState
*bs
)
2284 BDRVQcow2State
*s
= bs
->opaque
;
2285 qemu_vfree(s
->l1_table
);
2286 /* else pre-write overlap checks in cache_destroy may crash */
2289 if (!(s
->flags
& BDRV_O_INACTIVE
)) {
2290 qcow2_inactivate(bs
);
2293 cache_clean_timer_del(bs
);
2294 qcow2_cache_destroy(s
->l2_table_cache
);
2295 qcow2_cache_destroy(s
->refcount_block_cache
);
2297 qcrypto_block_free(s
->crypto
);
2300 g_free(s
->unknown_header_fields
);
2301 cleanup_unknown_header_ext(bs
);
2303 g_free(s
->image_data_file
);
2304 g_free(s
->image_backing_file
);
2305 g_free(s
->image_backing_format
);
2307 if (has_data_file(bs
)) {
2308 bdrv_unref_child(bs
, s
->data_file
);
2311 qcow2_refcount_close(bs
);
2312 qcow2_free_snapshots(bs
);
2315 static void coroutine_fn
qcow2_co_invalidate_cache(BlockDriverState
*bs
,
2318 BDRVQcow2State
*s
= bs
->opaque
;
2319 int flags
= s
->flags
;
2320 QCryptoBlock
*crypto
= NULL
;
2322 Error
*local_err
= NULL
;
2326 * Backing files are read-only which makes all of their metadata immutable,
2327 * that means we don't have to worry about reopening them here.
2335 memset(s
, 0, sizeof(BDRVQcow2State
));
2336 options
= qdict_clone_shallow(bs
->options
);
2338 flags
&= ~BDRV_O_INACTIVE
;
2339 qemu_co_mutex_lock(&s
->lock
);
2340 ret
= qcow2_do_open(bs
, options
, flags
, &local_err
);
2341 qemu_co_mutex_unlock(&s
->lock
);
2342 qobject_unref(options
);
2344 error_propagate_prepend(errp
, local_err
,
2345 "Could not reopen qcow2 layer: ");
2348 } else if (ret
< 0) {
2349 error_setg_errno(errp
, -ret
, "Could not reopen qcow2 layer");
2357 static size_t header_ext_add(char *buf
, uint32_t magic
, const void *s
,
2358 size_t len
, size_t buflen
)
2360 QCowExtension
*ext_backing_fmt
= (QCowExtension
*) buf
;
2361 size_t ext_len
= sizeof(QCowExtension
) + ((len
+ 7) & ~7);
2363 if (buflen
< ext_len
) {
2367 *ext_backing_fmt
= (QCowExtension
) {
2368 .magic
= cpu_to_be32(magic
),
2369 .len
= cpu_to_be32(len
),
2373 memcpy(buf
+ sizeof(QCowExtension
), s
, len
);
2380 * Updates the qcow2 header, including the variable length parts of it, i.e.
2381 * the backing file name and all extensions. qcow2 was not designed to allow
2382 * such changes, so if we run out of space (we can only use the first cluster)
2383 * this function may fail.
2385 * Returns 0 on success, -errno in error cases.
2387 int qcow2_update_header(BlockDriverState
*bs
)
2389 BDRVQcow2State
*s
= bs
->opaque
;
2392 size_t buflen
= s
->cluster_size
;
2394 uint64_t total_size
;
2395 uint32_t refcount_table_clusters
;
2396 size_t header_length
;
2397 Qcow2UnknownHeaderExtension
*uext
;
2399 buf
= qemu_blockalign(bs
, buflen
);
2401 /* Header structure */
2402 header
= (QCowHeader
*) buf
;
2404 if (buflen
< sizeof(*header
)) {
2409 header_length
= sizeof(*header
) + s
->unknown_header_fields_size
;
2410 total_size
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
2411 refcount_table_clusters
= s
->refcount_table_size
>> (s
->cluster_bits
- 3);
2413 *header
= (QCowHeader
) {
2414 /* Version 2 fields */
2415 .magic
= cpu_to_be32(QCOW_MAGIC
),
2416 .version
= cpu_to_be32(s
->qcow_version
),
2417 .backing_file_offset
= 0,
2418 .backing_file_size
= 0,
2419 .cluster_bits
= cpu_to_be32(s
->cluster_bits
),
2420 .size
= cpu_to_be64(total_size
),
2421 .crypt_method
= cpu_to_be32(s
->crypt_method_header
),
2422 .l1_size
= cpu_to_be32(s
->l1_size
),
2423 .l1_table_offset
= cpu_to_be64(s
->l1_table_offset
),
2424 .refcount_table_offset
= cpu_to_be64(s
->refcount_table_offset
),
2425 .refcount_table_clusters
= cpu_to_be32(refcount_table_clusters
),
2426 .nb_snapshots
= cpu_to_be32(s
->nb_snapshots
),
2427 .snapshots_offset
= cpu_to_be64(s
->snapshots_offset
),
2429 /* Version 3 fields */
2430 .incompatible_features
= cpu_to_be64(s
->incompatible_features
),
2431 .compatible_features
= cpu_to_be64(s
->compatible_features
),
2432 .autoclear_features
= cpu_to_be64(s
->autoclear_features
),
2433 .refcount_order
= cpu_to_be32(s
->refcount_order
),
2434 .header_length
= cpu_to_be32(header_length
),
2437 /* For older versions, write a shorter header */
2438 switch (s
->qcow_version
) {
2440 ret
= offsetof(QCowHeader
, incompatible_features
);
2443 ret
= sizeof(*header
);
2452 memset(buf
, 0, buflen
);
2454 /* Preserve any unknown field in the header */
2455 if (s
->unknown_header_fields_size
) {
2456 if (buflen
< s
->unknown_header_fields_size
) {
2461 memcpy(buf
, s
->unknown_header_fields
, s
->unknown_header_fields_size
);
2462 buf
+= s
->unknown_header_fields_size
;
2463 buflen
-= s
->unknown_header_fields_size
;
2466 /* Backing file format header extension */
2467 if (s
->image_backing_format
) {
2468 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BACKING_FORMAT
,
2469 s
->image_backing_format
,
2470 strlen(s
->image_backing_format
),
2480 /* External data file header extension */
2481 if (has_data_file(bs
) && s
->image_data_file
) {
2482 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_DATA_FILE
,
2483 s
->image_data_file
, strlen(s
->image_data_file
),
2493 /* Full disk encryption header pointer extension */
2494 if (s
->crypto_header
.offset
!= 0) {
2495 s
->crypto_header
.offset
= cpu_to_be64(s
->crypto_header
.offset
);
2496 s
->crypto_header
.length
= cpu_to_be64(s
->crypto_header
.length
);
2497 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_CRYPTO_HEADER
,
2498 &s
->crypto_header
, sizeof(s
->crypto_header
),
2500 s
->crypto_header
.offset
= be64_to_cpu(s
->crypto_header
.offset
);
2501 s
->crypto_header
.length
= be64_to_cpu(s
->crypto_header
.length
);
2510 if (s
->qcow_version
>= 3) {
2511 Qcow2Feature features
[] = {
2513 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2514 .bit
= QCOW2_INCOMPAT_DIRTY_BITNR
,
2515 .name
= "dirty bit",
2518 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2519 .bit
= QCOW2_INCOMPAT_CORRUPT_BITNR
,
2520 .name
= "corrupt bit",
2523 .type
= QCOW2_FEAT_TYPE_INCOMPATIBLE
,
2524 .bit
= QCOW2_INCOMPAT_DATA_FILE_BITNR
,
2525 .name
= "external data file",
2528 .type
= QCOW2_FEAT_TYPE_COMPATIBLE
,
2529 .bit
= QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR
,
2530 .name
= "lazy refcounts",
2534 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_FEATURE_TABLE
,
2535 features
, sizeof(features
), buflen
);
2543 /* Bitmap extension */
2544 if (s
->nb_bitmaps
> 0) {
2545 Qcow2BitmapHeaderExt bitmaps_header
= {
2546 .nb_bitmaps
= cpu_to_be32(s
->nb_bitmaps
),
2547 .bitmap_directory_size
=
2548 cpu_to_be64(s
->bitmap_directory_size
),
2549 .bitmap_directory_offset
=
2550 cpu_to_be64(s
->bitmap_directory_offset
)
2552 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_BITMAPS
,
2553 &bitmaps_header
, sizeof(bitmaps_header
),
2562 /* Keep unknown header extensions */
2563 QLIST_FOREACH(uext
, &s
->unknown_header_ext
, next
) {
2564 ret
= header_ext_add(buf
, uext
->magic
, uext
->data
, uext
->len
, buflen
);
2573 /* End of header extensions */
2574 ret
= header_ext_add(buf
, QCOW2_EXT_MAGIC_END
, NULL
, 0, buflen
);
2582 /* Backing file name */
2583 if (s
->image_backing_file
) {
2584 size_t backing_file_len
= strlen(s
->image_backing_file
);
2586 if (buflen
< backing_file_len
) {
2591 /* Using strncpy is ok here, since buf is not NUL-terminated. */
2592 strncpy(buf
, s
->image_backing_file
, buflen
);
2594 header
->backing_file_offset
= cpu_to_be64(buf
- ((char*) header
));
2595 header
->backing_file_size
= cpu_to_be32(backing_file_len
);
2598 /* Write the new header */
2599 ret
= bdrv_pwrite(bs
->file
, 0, header
, s
->cluster_size
);
2610 static int qcow2_change_backing_file(BlockDriverState
*bs
,
2611 const char *backing_file
, const char *backing_fmt
)
2613 BDRVQcow2State
*s
= bs
->opaque
;
2615 /* Adding a backing file means that the external data file alone won't be
2616 * enough to make sense of the content */
2617 if (backing_file
&& data_file_is_raw(bs
)) {
2621 if (backing_file
&& strlen(backing_file
) > 1023) {
2625 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2626 backing_file
?: "");
2627 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2628 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2630 g_free(s
->image_backing_file
);
2631 g_free(s
->image_backing_format
);
2633 s
->image_backing_file
= backing_file
? g_strdup(bs
->backing_file
) : NULL
;
2634 s
->image_backing_format
= backing_fmt
? g_strdup(bs
->backing_format
) : NULL
;
2636 return qcow2_update_header(bs
);
2639 static int qcow2_crypt_method_from_format(const char *encryptfmt
)
2641 if (g_str_equal(encryptfmt
, "luks")) {
2642 return QCOW_CRYPT_LUKS
;
2643 } else if (g_str_equal(encryptfmt
, "aes")) {
2644 return QCOW_CRYPT_AES
;
2650 static int qcow2_set_up_encryption(BlockDriverState
*bs
,
2651 QCryptoBlockCreateOptions
*cryptoopts
,
2654 BDRVQcow2State
*s
= bs
->opaque
;
2655 QCryptoBlock
*crypto
= NULL
;
2658 switch (cryptoopts
->format
) {
2659 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
2660 fmt
= QCOW_CRYPT_LUKS
;
2662 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
2663 fmt
= QCOW_CRYPT_AES
;
2666 error_setg(errp
, "Crypto format not supported in qcow2");
2670 s
->crypt_method_header
= fmt
;
2672 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
2673 qcow2_crypto_hdr_init_func
,
2674 qcow2_crypto_hdr_write_func
,
2680 ret
= qcow2_update_header(bs
);
2682 error_setg_errno(errp
, -ret
, "Could not write encryption header");
2688 qcrypto_block_free(crypto
);
2693 * Preallocates metadata structures for data clusters between @offset (in the
2694 * guest disk) and @new_length (which is thus generally the new guest disk
2697 * Returns: 0 on success, -errno on failure.
2699 static int coroutine_fn
preallocate_co(BlockDriverState
*bs
, uint64_t offset
,
2700 uint64_t new_length
)
2703 uint64_t host_offset
= 0;
2704 unsigned int cur_bytes
;
2708 assert(offset
<= new_length
);
2709 bytes
= new_length
- offset
;
2712 cur_bytes
= MIN(bytes
, INT_MAX
);
2713 ret
= qcow2_alloc_cluster_offset(bs
, offset
, &cur_bytes
,
2714 &host_offset
, &meta
);
2720 QCowL2Meta
*next
= meta
->next
;
2722 ret
= qcow2_alloc_cluster_link_l2(bs
, meta
);
2724 qcow2_free_any_clusters(bs
, meta
->alloc_offset
,
2725 meta
->nb_clusters
, QCOW2_DISCARD_NEVER
);
2729 /* There are no dependent requests, but we need to remove our
2730 * request from the list of in-flight requests */
2731 QLIST_REMOVE(meta
, next_in_flight
);
2737 /* TODO Preallocate data if requested */
2740 offset
+= cur_bytes
;
2744 * It is expected that the image file is large enough to actually contain
2745 * all of the allocated clusters (otherwise we get failing reads after
2746 * EOF). Extend the image to the last allocated sector.
2748 if (host_offset
!= 0) {
2750 ret
= bdrv_pwrite(bs
->file
, (host_offset
+ cur_bytes
) - 1,
2760 /* qcow2_refcount_metadata_size:
2761 * @clusters: number of clusters to refcount (including data and L1/L2 tables)
2762 * @cluster_size: size of a cluster, in bytes
2763 * @refcount_order: refcount bits power-of-2 exponent
2764 * @generous_increase: allow for the refcount table to be 1.5x as large as it
2767 * Returns: Number of bytes required for refcount blocks and table metadata.
2769 int64_t qcow2_refcount_metadata_size(int64_t clusters
, size_t cluster_size
,
2770 int refcount_order
, bool generous_increase
,
2771 uint64_t *refblock_count
)
2774 * Every host cluster is reference-counted, including metadata (even
2775 * refcount metadata is recursively included).
2777 * An accurate formula for the size of refcount metadata size is difficult
2778 * to derive. An easier method of calculation is finding the fixed point
2779 * where no further refcount blocks or table clusters are required to
2780 * reference count every cluster.
2782 int64_t blocks_per_table_cluster
= cluster_size
/ sizeof(uint64_t);
2783 int64_t refcounts_per_block
= cluster_size
* 8 / (1 << refcount_order
);
2784 int64_t table
= 0; /* number of refcount table clusters */
2785 int64_t blocks
= 0; /* number of refcount block clusters */
2791 blocks
= DIV_ROUND_UP(clusters
+ table
+ blocks
, refcounts_per_block
);
2792 table
= DIV_ROUND_UP(blocks
, blocks_per_table_cluster
);
2793 n
= clusters
+ blocks
+ table
;
2795 if (n
== last
&& generous_increase
) {
2796 clusters
+= DIV_ROUND_UP(table
, 2);
2797 n
= 0; /* force another loop */
2798 generous_increase
= false;
2800 } while (n
!= last
);
2802 if (refblock_count
) {
2803 *refblock_count
= blocks
;
2806 return (blocks
+ table
) * cluster_size
;
2810 * qcow2_calc_prealloc_size:
2811 * @total_size: virtual disk size in bytes
2812 * @cluster_size: cluster size in bytes
2813 * @refcount_order: refcount bits power-of-2 exponent
2815 * Returns: Total number of bytes required for the fully allocated image
2816 * (including metadata).
2818 static int64_t qcow2_calc_prealloc_size(int64_t total_size
,
2819 size_t cluster_size
,
2822 int64_t meta_size
= 0;
2823 uint64_t nl1e
, nl2e
;
2824 int64_t aligned_total_size
= ROUND_UP(total_size
, cluster_size
);
2826 /* header: 1 cluster */
2827 meta_size
+= cluster_size
;
2829 /* total size of L2 tables */
2830 nl2e
= aligned_total_size
/ cluster_size
;
2831 nl2e
= ROUND_UP(nl2e
, cluster_size
/ sizeof(uint64_t));
2832 meta_size
+= nl2e
* sizeof(uint64_t);
2834 /* total size of L1 tables */
2835 nl1e
= nl2e
* sizeof(uint64_t) / cluster_size
;
2836 nl1e
= ROUND_UP(nl1e
, cluster_size
/ sizeof(uint64_t));
2837 meta_size
+= nl1e
* sizeof(uint64_t);
2839 /* total size of refcount table and blocks */
2840 meta_size
+= qcow2_refcount_metadata_size(
2841 (meta_size
+ aligned_total_size
) / cluster_size
,
2842 cluster_size
, refcount_order
, false, NULL
);
2844 return meta_size
+ aligned_total_size
;
2847 static bool validate_cluster_size(size_t cluster_size
, Error
**errp
)
2849 int cluster_bits
= ctz32(cluster_size
);
2850 if (cluster_bits
< MIN_CLUSTER_BITS
|| cluster_bits
> MAX_CLUSTER_BITS
||
2851 (1 << cluster_bits
) != cluster_size
)
2853 error_setg(errp
, "Cluster size must be a power of two between %d and "
2854 "%dk", 1 << MIN_CLUSTER_BITS
, 1 << (MAX_CLUSTER_BITS
- 10));
2860 static size_t qcow2_opt_get_cluster_size_del(QemuOpts
*opts
, Error
**errp
)
2862 size_t cluster_size
;
2864 cluster_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_CLUSTER_SIZE
,
2865 DEFAULT_CLUSTER_SIZE
);
2866 if (!validate_cluster_size(cluster_size
, errp
)) {
2869 return cluster_size
;
2872 static int qcow2_opt_get_version_del(QemuOpts
*opts
, Error
**errp
)
2877 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_COMPAT_LEVEL
);
2879 ret
= 3; /* default */
2880 } else if (!strcmp(buf
, "0.10")) {
2882 } else if (!strcmp(buf
, "1.1")) {
2885 error_setg(errp
, "Invalid compatibility level: '%s'", buf
);
2892 static uint64_t qcow2_opt_get_refcount_bits_del(QemuOpts
*opts
, int version
,
2895 uint64_t refcount_bits
;
2897 refcount_bits
= qemu_opt_get_number_del(opts
, BLOCK_OPT_REFCOUNT_BITS
, 16);
2898 if (refcount_bits
> 64 || !is_power_of_2(refcount_bits
)) {
2899 error_setg(errp
, "Refcount width must be a power of two and may not "
2904 if (version
< 3 && refcount_bits
!= 16) {
2905 error_setg(errp
, "Different refcount widths than 16 bits require "
2906 "compatibility level 1.1 or above (use compat=1.1 or "
2911 return refcount_bits
;
2914 static int coroutine_fn
2915 qcow2_co_create(BlockdevCreateOptions
*create_options
, Error
**errp
)
2917 BlockdevCreateOptionsQcow2
*qcow2_opts
;
2921 * Open the image file and write a minimal qcow2 header.
2923 * We keep things simple and start with a zero-sized image. We also
2924 * do without refcount blocks or a L1 table for now. We'll fix the
2925 * inconsistency later.
2927 * We do need a refcount table because growing the refcount table means
2928 * allocating two new refcount blocks - the seconds of which would be at
2929 * 2 GB for 64k clusters, and we don't want to have a 2 GB initial file
2930 * size for any qcow2 image.
2932 BlockBackend
*blk
= NULL
;
2933 BlockDriverState
*bs
= NULL
;
2934 BlockDriverState
*data_bs
= NULL
;
2936 size_t cluster_size
;
2939 uint64_t* refcount_table
;
2940 Error
*local_err
= NULL
;
2943 assert(create_options
->driver
== BLOCKDEV_DRIVER_QCOW2
);
2944 qcow2_opts
= &create_options
->u
.qcow2
;
2946 bs
= bdrv_open_blockdev_ref(qcow2_opts
->file
, errp
);
2951 /* Validate options and set default values */
2952 if (!QEMU_IS_ALIGNED(qcow2_opts
->size
, BDRV_SECTOR_SIZE
)) {
2953 error_setg(errp
, "Image size must be a multiple of 512 bytes");
2958 if (qcow2_opts
->has_version
) {
2959 switch (qcow2_opts
->version
) {
2960 case BLOCKDEV_QCOW2_VERSION_V2
:
2963 case BLOCKDEV_QCOW2_VERSION_V3
:
2967 g_assert_not_reached();
2973 if (qcow2_opts
->has_cluster_size
) {
2974 cluster_size
= qcow2_opts
->cluster_size
;
2976 cluster_size
= DEFAULT_CLUSTER_SIZE
;
2979 if (!validate_cluster_size(cluster_size
, errp
)) {
2984 if (!qcow2_opts
->has_preallocation
) {
2985 qcow2_opts
->preallocation
= PREALLOC_MODE_OFF
;
2987 if (qcow2_opts
->has_backing_file
&&
2988 qcow2_opts
->preallocation
!= PREALLOC_MODE_OFF
)
2990 error_setg(errp
, "Backing file and preallocation cannot be used at "
2995 if (qcow2_opts
->has_backing_fmt
&& !qcow2_opts
->has_backing_file
) {
2996 error_setg(errp
, "Backing format cannot be used without backing file");
3001 if (!qcow2_opts
->has_lazy_refcounts
) {
3002 qcow2_opts
->lazy_refcounts
= false;
3004 if (version
< 3 && qcow2_opts
->lazy_refcounts
) {
3005 error_setg(errp
, "Lazy refcounts only supported with compatibility "
3006 "level 1.1 and above (use version=v3 or greater)");
3011 if (!qcow2_opts
->has_refcount_bits
) {
3012 qcow2_opts
->refcount_bits
= 16;
3014 if (qcow2_opts
->refcount_bits
> 64 ||
3015 !is_power_of_2(qcow2_opts
->refcount_bits
))
3017 error_setg(errp
, "Refcount width must be a power of two and may not "
3022 if (version
< 3 && qcow2_opts
->refcount_bits
!= 16) {
3023 error_setg(errp
, "Different refcount widths than 16 bits require "
3024 "compatibility level 1.1 or above (use version=v3 or "
3029 refcount_order
= ctz32(qcow2_opts
->refcount_bits
);
3031 if (qcow2_opts
->data_file_raw
&& !qcow2_opts
->data_file
) {
3032 error_setg(errp
, "data-file-raw requires data-file");
3036 if (qcow2_opts
->data_file_raw
&& qcow2_opts
->has_backing_file
) {
3037 error_setg(errp
, "Backing file and data-file-raw cannot be used at "
3043 if (qcow2_opts
->data_file
) {
3045 error_setg(errp
, "External data files are only supported with "
3046 "compatibility level 1.1 and above (use version=v3 or "
3051 data_bs
= bdrv_open_blockdev_ref(qcow2_opts
->data_file
, errp
);
3058 /* Create BlockBackend to write to the image */
3059 blk
= blk_new(BLK_PERM_WRITE
| BLK_PERM_RESIZE
, BLK_PERM_ALL
);
3060 ret
= blk_insert_bs(blk
, bs
, errp
);
3064 blk_set_allow_write_beyond_eof(blk
, true);
3066 /* Clear the protocol layer and preallocate it if necessary */
3067 ret
= blk_truncate(blk
, 0, PREALLOC_MODE_OFF
, errp
);
3072 /* Write the header */
3073 QEMU_BUILD_BUG_ON((1 << MIN_CLUSTER_BITS
) < sizeof(*header
));
3074 header
= g_malloc0(cluster_size
);
3075 *header
= (QCowHeader
) {
3076 .magic
= cpu_to_be32(QCOW_MAGIC
),
3077 .version
= cpu_to_be32(version
),
3078 .cluster_bits
= cpu_to_be32(ctz32(cluster_size
)),
3079 .size
= cpu_to_be64(0),
3080 .l1_table_offset
= cpu_to_be64(0),
3081 .l1_size
= cpu_to_be32(0),
3082 .refcount_table_offset
= cpu_to_be64(cluster_size
),
3083 .refcount_table_clusters
= cpu_to_be32(1),
3084 .refcount_order
= cpu_to_be32(refcount_order
),
3085 .header_length
= cpu_to_be32(sizeof(*header
)),
3088 /* We'll update this to correct value later */
3089 header
->crypt_method
= cpu_to_be32(QCOW_CRYPT_NONE
);
3091 if (qcow2_opts
->lazy_refcounts
) {
3092 header
->compatible_features
|=
3093 cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS
);
3096 header
->incompatible_features
|=
3097 cpu_to_be64(QCOW2_INCOMPAT_DATA_FILE
);
3099 if (qcow2_opts
->data_file_raw
) {
3100 header
->autoclear_features
|=
3101 cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW
);
3104 ret
= blk_pwrite(blk
, 0, header
, cluster_size
, 0);
3107 error_setg_errno(errp
, -ret
, "Could not write qcow2 header");
3111 /* Write a refcount table with one refcount block */
3112 refcount_table
= g_malloc0(2 * cluster_size
);
3113 refcount_table
[0] = cpu_to_be64(2 * cluster_size
);
3114 ret
= blk_pwrite(blk
, cluster_size
, refcount_table
, 2 * cluster_size
, 0);
3115 g_free(refcount_table
);
3118 error_setg_errno(errp
, -ret
, "Could not write refcount table");
3126 * And now open the image and make it consistent first (i.e. increase the
3127 * refcount of the cluster that is occupied by the header and the refcount
3130 options
= qdict_new();
3131 qdict_put_str(options
, "driver", "qcow2");
3132 qdict_put_str(options
, "file", bs
->node_name
);
3134 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3136 blk
= blk_new_open(NULL
, NULL
, options
,
3137 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_NO_FLUSH
,
3140 error_propagate(errp
, local_err
);
3145 ret
= qcow2_alloc_clusters(blk_bs(blk
), 3 * cluster_size
);
3147 error_setg_errno(errp
, -ret
, "Could not allocate clusters for qcow2 "
3148 "header and refcount table");
3151 } else if (ret
!= 0) {
3152 error_report("Huh, first cluster in empty image is already in use?");
3156 /* Set the external data file if necessary */
3158 BDRVQcow2State
*s
= blk_bs(blk
)->opaque
;
3159 s
->image_data_file
= g_strdup(data_bs
->filename
);
3162 /* Create a full header (including things like feature table) */
3163 ret
= qcow2_update_header(blk_bs(blk
));
3165 error_setg_errno(errp
, -ret
, "Could not update qcow2 header");
3169 /* Okay, now that we have a valid image, let's give it the right size */
3170 ret
= blk_truncate(blk
, qcow2_opts
->size
, qcow2_opts
->preallocation
, errp
);
3172 error_prepend(errp
, "Could not resize image: ");
3176 /* Want a backing file? There you go.*/
3177 if (qcow2_opts
->has_backing_file
) {
3178 const char *backing_format
= NULL
;
3180 if (qcow2_opts
->has_backing_fmt
) {
3181 backing_format
= BlockdevDriver_str(qcow2_opts
->backing_fmt
);
3184 ret
= bdrv_change_backing_file(blk_bs(blk
), qcow2_opts
->backing_file
,
3187 error_setg_errno(errp
, -ret
, "Could not assign backing file '%s' "
3188 "with format '%s'", qcow2_opts
->backing_file
,
3194 /* Want encryption? There you go. */
3195 if (qcow2_opts
->has_encrypt
) {
3196 ret
= qcow2_set_up_encryption(blk_bs(blk
), qcow2_opts
->encrypt
, errp
);
3205 /* Reopen the image without BDRV_O_NO_FLUSH to flush it before returning.
3206 * Using BDRV_O_NO_IO, since encryption is now setup we don't want to
3207 * have to setup decryption context. We're not doing any I/O on the top
3208 * level BlockDriverState, only lower layers, where BDRV_O_NO_IO does
3211 options
= qdict_new();
3212 qdict_put_str(options
, "driver", "qcow2");
3213 qdict_put_str(options
, "file", bs
->node_name
);
3215 qdict_put_str(options
, "data-file", data_bs
->node_name
);
3217 blk
= blk_new_open(NULL
, NULL
, options
,
3218 BDRV_O_RDWR
| BDRV_O_NO_BACKING
| BDRV_O_NO_IO
,
3221 error_propagate(errp
, local_err
);
3230 bdrv_unref(data_bs
);
3234 static int coroutine_fn
qcow2_co_create_opts(const char *filename
, QemuOpts
*opts
,
3237 BlockdevCreateOptions
*create_options
= NULL
;
3240 BlockDriverState
*bs
= NULL
;
3241 BlockDriverState
*data_bs
= NULL
;
3242 Error
*local_err
= NULL
;
3246 /* Only the keyval visitor supports the dotted syntax needed for
3247 * encryption, so go through a QDict before getting a QAPI type. Ignore
3248 * options meant for the protocol layer so that the visitor doesn't
3250 qdict
= qemu_opts_to_qdict_filtered(opts
, NULL
, bdrv_qcow2
.create_opts
,
3253 /* Handle encryption options */
3254 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT
);
3255 if (val
&& !strcmp(val
, "on")) {
3256 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT
, "qcow");
3257 } else if (val
&& !strcmp(val
, "off")) {
3258 qdict_del(qdict
, BLOCK_OPT_ENCRYPT
);
3261 val
= qdict_get_try_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
);
3262 if (val
&& !strcmp(val
, "aes")) {
3263 qdict_put_str(qdict
, BLOCK_OPT_ENCRYPT_FORMAT
, "qcow");
3266 /* Convert compat=0.10/1.1 into compat=v2/v3, to be renamed into
3267 * version=v2/v3 below. */
3268 val
= qdict_get_try_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
);
3269 if (val
&& !strcmp(val
, "0.10")) {
3270 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v2");
3271 } else if (val
&& !strcmp(val
, "1.1")) {
3272 qdict_put_str(qdict
, BLOCK_OPT_COMPAT_LEVEL
, "v3");
3275 /* Change legacy command line options into QMP ones */
3276 static const QDictRenames opt_renames
[] = {
3277 { BLOCK_OPT_BACKING_FILE
, "backing-file" },
3278 { BLOCK_OPT_BACKING_FMT
, "backing-fmt" },
3279 { BLOCK_OPT_CLUSTER_SIZE
, "cluster-size" },
3280 { BLOCK_OPT_LAZY_REFCOUNTS
, "lazy-refcounts" },
3281 { BLOCK_OPT_REFCOUNT_BITS
, "refcount-bits" },
3282 { BLOCK_OPT_ENCRYPT
, BLOCK_OPT_ENCRYPT_FORMAT
},
3283 { BLOCK_OPT_COMPAT_LEVEL
, "version" },
3284 { BLOCK_OPT_DATA_FILE_RAW
, "data-file-raw" },
3288 if (!qdict_rename_keys(qdict
, opt_renames
, errp
)) {
3293 /* Create and open the file (protocol layer) */
3294 ret
= bdrv_create_file(filename
, opts
, errp
);
3299 bs
= bdrv_open(filename
, NULL
, NULL
,
3300 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
, errp
);
3306 /* Create and open an external data file (protocol layer) */
3307 val
= qdict_get_try_str(qdict
, BLOCK_OPT_DATA_FILE
);
3309 ret
= bdrv_create_file(val
, opts
, errp
);
3314 data_bs
= bdrv_open(val
, NULL
, NULL
,
3315 BDRV_O_RDWR
| BDRV_O_RESIZE
| BDRV_O_PROTOCOL
,
3317 if (data_bs
== NULL
) {
3322 qdict_del(qdict
, BLOCK_OPT_DATA_FILE
);
3323 qdict_put_str(qdict
, "data-file", data_bs
->node_name
);
3326 /* Set 'driver' and 'node' options */
3327 qdict_put_str(qdict
, "driver", "qcow2");
3328 qdict_put_str(qdict
, "file", bs
->node_name
);
3330 /* Now get the QAPI type BlockdevCreateOptions */
3331 v
= qobject_input_visitor_new_flat_confused(qdict
, errp
);
3337 visit_type_BlockdevCreateOptions(v
, NULL
, &create_options
, &local_err
);
3341 error_propagate(errp
, local_err
);
3346 /* Silently round up size */
3347 create_options
->u
.qcow2
.size
= ROUND_UP(create_options
->u
.qcow2
.size
,
3350 /* Create the qcow2 image (format layer) */
3351 ret
= qcow2_co_create(create_options
, errp
);
3358 qobject_unref(qdict
);
3360 bdrv_unref(data_bs
);
3361 qapi_free_BlockdevCreateOptions(create_options
);
3366 static bool is_zero(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3371 /* Clamp to image length, before checking status of underlying sectors */
3372 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3373 bytes
= bs
->total_sectors
* BDRV_SECTOR_SIZE
- offset
;
3379 res
= bdrv_block_status_above(bs
, NULL
, offset
, bytes
, &nr
, NULL
, NULL
);
3380 return res
>= 0 && (res
& BDRV_BLOCK_ZERO
) && nr
== bytes
;
3383 static coroutine_fn
int qcow2_co_pwrite_zeroes(BlockDriverState
*bs
,
3384 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3387 BDRVQcow2State
*s
= bs
->opaque
;
3389 uint32_t head
= offset
% s
->cluster_size
;
3390 uint32_t tail
= (offset
+ bytes
) % s
->cluster_size
;
3392 trace_qcow2_pwrite_zeroes_start_req(qemu_coroutine_self(), offset
, bytes
);
3393 if (offset
+ bytes
== bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3401 assert(head
+ bytes
<= s
->cluster_size
);
3403 /* check whether remainder of cluster already reads as zero */
3404 if (!(is_zero(bs
, offset
- head
, head
) &&
3405 is_zero(bs
, offset
+ bytes
,
3406 tail
? s
->cluster_size
- tail
: 0))) {
3410 qemu_co_mutex_lock(&s
->lock
);
3411 /* We can have new write after previous check */
3412 offset
= QEMU_ALIGN_DOWN(offset
, s
->cluster_size
);
3413 bytes
= s
->cluster_size
;
3414 nr
= s
->cluster_size
;
3415 ret
= qcow2_get_cluster_offset(bs
, offset
, &nr
, &off
);
3416 if (ret
!= QCOW2_CLUSTER_UNALLOCATED
&&
3417 ret
!= QCOW2_CLUSTER_ZERO_PLAIN
&&
3418 ret
!= QCOW2_CLUSTER_ZERO_ALLOC
) {
3419 qemu_co_mutex_unlock(&s
->lock
);
3423 qemu_co_mutex_lock(&s
->lock
);
3426 trace_qcow2_pwrite_zeroes(qemu_coroutine_self(), offset
, bytes
);
3428 /* Whatever is left can use real zero clusters */
3429 ret
= qcow2_cluster_zeroize(bs
, offset
, bytes
, flags
);
3430 qemu_co_mutex_unlock(&s
->lock
);
3435 static coroutine_fn
int qcow2_co_pdiscard(BlockDriverState
*bs
,
3436 int64_t offset
, int bytes
)
3439 BDRVQcow2State
*s
= bs
->opaque
;
3441 if (!QEMU_IS_ALIGNED(offset
| bytes
, s
->cluster_size
)) {
3442 assert(bytes
< s
->cluster_size
);
3443 /* Ignore partial clusters, except for the special case of the
3444 * complete partial cluster at the end of an unaligned file */
3445 if (!QEMU_IS_ALIGNED(offset
, s
->cluster_size
) ||
3446 offset
+ bytes
!= bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3451 qemu_co_mutex_lock(&s
->lock
);
3452 ret
= qcow2_cluster_discard(bs
, offset
, bytes
, QCOW2_DISCARD_REQUEST
,
3454 qemu_co_mutex_unlock(&s
->lock
);
3458 static int coroutine_fn
3459 qcow2_co_copy_range_from(BlockDriverState
*bs
,
3460 BdrvChild
*src
, uint64_t src_offset
,
3461 BdrvChild
*dst
, uint64_t dst_offset
,
3462 uint64_t bytes
, BdrvRequestFlags read_flags
,
3463 BdrvRequestFlags write_flags
)
3465 BDRVQcow2State
*s
= bs
->opaque
;
3467 unsigned int cur_bytes
; /* number of bytes in current iteration */
3468 BdrvChild
*child
= NULL
;
3469 BdrvRequestFlags cur_write_flags
;
3471 assert(!bs
->encrypted
);
3472 qemu_co_mutex_lock(&s
->lock
);
3474 while (bytes
!= 0) {
3475 uint64_t copy_offset
= 0;
3476 /* prepare next request */
3477 cur_bytes
= MIN(bytes
, INT_MAX
);
3478 cur_write_flags
= write_flags
;
3480 ret
= qcow2_get_cluster_offset(bs
, src_offset
, &cur_bytes
, ©_offset
);
3486 case QCOW2_CLUSTER_UNALLOCATED
:
3487 if (bs
->backing
&& bs
->backing
->bs
) {
3488 int64_t backing_length
= bdrv_getlength(bs
->backing
->bs
);
3489 if (src_offset
>= backing_length
) {
3490 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3492 child
= bs
->backing
;
3493 cur_bytes
= MIN(cur_bytes
, backing_length
- src_offset
);
3494 copy_offset
= src_offset
;
3497 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3501 case QCOW2_CLUSTER_ZERO_PLAIN
:
3502 case QCOW2_CLUSTER_ZERO_ALLOC
:
3503 cur_write_flags
|= BDRV_REQ_ZERO_WRITE
;
3506 case QCOW2_CLUSTER_COMPRESSED
:
3510 case QCOW2_CLUSTER_NORMAL
:
3511 child
= s
->data_file
;
3512 copy_offset
+= offset_into_cluster(s
, src_offset
);
3513 if ((copy_offset
& 511) != 0) {
3522 qemu_co_mutex_unlock(&s
->lock
);
3523 ret
= bdrv_co_copy_range_from(child
,
3526 cur_bytes
, read_flags
, cur_write_flags
);
3527 qemu_co_mutex_lock(&s
->lock
);
3533 src_offset
+= cur_bytes
;
3534 dst_offset
+= cur_bytes
;
3539 qemu_co_mutex_unlock(&s
->lock
);
3543 static int coroutine_fn
3544 qcow2_co_copy_range_to(BlockDriverState
*bs
,
3545 BdrvChild
*src
, uint64_t src_offset
,
3546 BdrvChild
*dst
, uint64_t dst_offset
,
3547 uint64_t bytes
, BdrvRequestFlags read_flags
,
3548 BdrvRequestFlags write_flags
)
3550 BDRVQcow2State
*s
= bs
->opaque
;
3551 int offset_in_cluster
;
3553 unsigned int cur_bytes
; /* number of sectors in current iteration */
3554 uint64_t cluster_offset
;
3555 QCowL2Meta
*l2meta
= NULL
;
3557 assert(!bs
->encrypted
);
3559 qemu_co_mutex_lock(&s
->lock
);
3561 while (bytes
!= 0) {
3565 offset_in_cluster
= offset_into_cluster(s
, dst_offset
);
3566 cur_bytes
= MIN(bytes
, INT_MAX
);
3569 * If src->bs == dst->bs, we could simply copy by incrementing
3570 * the refcnt, without copying user data.
3571 * Or if src->bs == dst->bs->backing->bs, we could copy by discarding. */
3572 ret
= qcow2_alloc_cluster_offset(bs
, dst_offset
, &cur_bytes
,
3573 &cluster_offset
, &l2meta
);
3578 assert((cluster_offset
& 511) == 0);
3580 ret
= qcow2_pre_write_overlap_check(bs
, 0,
3581 cluster_offset
+ offset_in_cluster
, cur_bytes
, true);
3586 qemu_co_mutex_unlock(&s
->lock
);
3587 ret
= bdrv_co_copy_range_to(src
, src_offset
,
3589 cluster_offset
+ offset_in_cluster
,
3590 cur_bytes
, read_flags
, write_flags
);
3591 qemu_co_mutex_lock(&s
->lock
);
3596 ret
= qcow2_handle_l2meta(bs
, &l2meta
, true);
3602 src_offset
+= cur_bytes
;
3603 dst_offset
+= cur_bytes
;
3608 qcow2_handle_l2meta(bs
, &l2meta
, false);
3610 qemu_co_mutex_unlock(&s
->lock
);
3612 trace_qcow2_writev_done_req(qemu_coroutine_self(), ret
);
3617 static int coroutine_fn
qcow2_co_truncate(BlockDriverState
*bs
, int64_t offset
,
3618 PreallocMode prealloc
, Error
**errp
)
3620 BDRVQcow2State
*s
= bs
->opaque
;
3621 uint64_t old_length
;
3622 int64_t new_l1_size
;
3626 if (prealloc
!= PREALLOC_MODE_OFF
&& prealloc
!= PREALLOC_MODE_METADATA
&&
3627 prealloc
!= PREALLOC_MODE_FALLOC
&& prealloc
!= PREALLOC_MODE_FULL
)
3629 error_setg(errp
, "Unsupported preallocation mode '%s'",
3630 PreallocMode_str(prealloc
));
3635 error_setg(errp
, "The new size must be a multiple of 512");
3639 qemu_co_mutex_lock(&s
->lock
);
3641 /* cannot proceed if image has snapshots */
3642 if (s
->nb_snapshots
) {
3643 error_setg(errp
, "Can't resize an image which has snapshots");
3648 /* cannot proceed if image has bitmaps */
3649 if (s
->nb_bitmaps
) {
3650 /* TODO: resize bitmaps in the image */
3651 error_setg(errp
, "Can't resize an image which has bitmaps");
3656 old_length
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3657 new_l1_size
= size_to_l1(s
, offset
);
3659 if (offset
< old_length
) {
3660 int64_t last_cluster
, old_file_size
;
3661 if (prealloc
!= PREALLOC_MODE_OFF
) {
3663 "Preallocation can't be used for shrinking an image");
3668 ret
= qcow2_cluster_discard(bs
, ROUND_UP(offset
, s
->cluster_size
),
3669 old_length
- ROUND_UP(offset
,
3671 QCOW2_DISCARD_ALWAYS
, true);
3673 error_setg_errno(errp
, -ret
, "Failed to discard cropped clusters");
3677 ret
= qcow2_shrink_l1_table(bs
, new_l1_size
);
3679 error_setg_errno(errp
, -ret
,
3680 "Failed to reduce the number of L2 tables");
3684 ret
= qcow2_shrink_reftable(bs
);
3686 error_setg_errno(errp
, -ret
,
3687 "Failed to discard unused refblocks");
3691 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3692 if (old_file_size
< 0) {
3693 error_setg_errno(errp
, -old_file_size
,
3694 "Failed to inquire current file length");
3695 ret
= old_file_size
;
3698 last_cluster
= qcow2_get_last_cluster(bs
, old_file_size
);
3699 if (last_cluster
< 0) {
3700 error_setg_errno(errp
, -last_cluster
,
3701 "Failed to find the last cluster");
3705 if ((last_cluster
+ 1) * s
->cluster_size
< old_file_size
) {
3706 Error
*local_err
= NULL
;
3708 bdrv_co_truncate(bs
->file
, (last_cluster
+ 1) * s
->cluster_size
,
3709 PREALLOC_MODE_OFF
, &local_err
);
3711 warn_reportf_err(local_err
,
3712 "Failed to truncate the tail of the image: ");
3716 ret
= qcow2_grow_l1_table(bs
, new_l1_size
, true);
3718 error_setg_errno(errp
, -ret
, "Failed to grow the L1 table");
3724 case PREALLOC_MODE_OFF
:
3727 case PREALLOC_MODE_METADATA
:
3728 ret
= preallocate_co(bs
, old_length
, offset
);
3730 error_setg_errno(errp
, -ret
, "Preallocation failed");
3735 case PREALLOC_MODE_FALLOC
:
3736 case PREALLOC_MODE_FULL
:
3738 int64_t allocation_start
, host_offset
, guest_offset
;
3739 int64_t clusters_allocated
;
3740 int64_t old_file_size
, new_file_size
;
3741 uint64_t nb_new_data_clusters
, nb_new_l2_tables
;
3743 /* With a data file, preallocation means just allocating the metadata
3744 * and forwarding the truncate request to the data file */
3745 if (has_data_file(bs
)) {
3746 ret
= preallocate_co(bs
, old_length
, offset
);
3748 error_setg_errno(errp
, -ret
, "Preallocation failed");
3754 old_file_size
= bdrv_getlength(bs
->file
->bs
);
3755 if (old_file_size
< 0) {
3756 error_setg_errno(errp
, -old_file_size
,
3757 "Failed to inquire current file length");
3758 ret
= old_file_size
;
3761 old_file_size
= ROUND_UP(old_file_size
, s
->cluster_size
);
3763 nb_new_data_clusters
= DIV_ROUND_UP(offset
- old_length
,
3766 /* This is an overestimation; we will not actually allocate space for
3767 * these in the file but just make sure the new refcount structures are
3768 * able to cover them so we will not have to allocate new refblocks
3769 * while entering the data blocks in the potentially new L2 tables.
3770 * (We do not actually care where the L2 tables are placed. Maybe they
3771 * are already allocated or they can be placed somewhere before
3772 * @old_file_size. It does not matter because they will be fully
3773 * allocated automatically, so they do not need to be covered by the
3774 * preallocation. All that matters is that we will not have to allocate
3775 * new refcount structures for them.) */
3776 nb_new_l2_tables
= DIV_ROUND_UP(nb_new_data_clusters
,
3777 s
->cluster_size
/ sizeof(uint64_t));
3778 /* The cluster range may not be aligned to L2 boundaries, so add one L2
3779 * table for a potential head/tail */
3782 allocation_start
= qcow2_refcount_area(bs
, old_file_size
,
3783 nb_new_data_clusters
+
3786 if (allocation_start
< 0) {
3787 error_setg_errno(errp
, -allocation_start
,
3788 "Failed to resize refcount structures");
3789 ret
= allocation_start
;
3793 clusters_allocated
= qcow2_alloc_clusters_at(bs
, allocation_start
,
3794 nb_new_data_clusters
);
3795 if (clusters_allocated
< 0) {
3796 error_setg_errno(errp
, -clusters_allocated
,
3797 "Failed to allocate data clusters");
3798 ret
= clusters_allocated
;
3802 assert(clusters_allocated
== nb_new_data_clusters
);
3804 /* Allocate the data area */
3805 new_file_size
= allocation_start
+
3806 nb_new_data_clusters
* s
->cluster_size
;
3807 ret
= bdrv_co_truncate(bs
->file
, new_file_size
, prealloc
, errp
);
3809 error_prepend(errp
, "Failed to resize underlying file: ");
3810 qcow2_free_clusters(bs
, allocation_start
,
3811 nb_new_data_clusters
* s
->cluster_size
,
3812 QCOW2_DISCARD_OTHER
);
3816 /* Create the necessary L2 entries */
3817 host_offset
= allocation_start
;
3818 guest_offset
= old_length
;
3819 while (nb_new_data_clusters
) {
3820 int64_t nb_clusters
= MIN(
3821 nb_new_data_clusters
,
3822 s
->l2_slice_size
- offset_to_l2_slice_index(s
, guest_offset
));
3823 QCowL2Meta allocation
= {
3824 .offset
= guest_offset
,
3825 .alloc_offset
= host_offset
,
3826 .nb_clusters
= nb_clusters
,
3828 qemu_co_queue_init(&allocation
.dependent_requests
);
3830 ret
= qcow2_alloc_cluster_link_l2(bs
, &allocation
);
3832 error_setg_errno(errp
, -ret
, "Failed to update L2 tables");
3833 qcow2_free_clusters(bs
, host_offset
,
3834 nb_new_data_clusters
* s
->cluster_size
,
3835 QCOW2_DISCARD_OTHER
);
3839 guest_offset
+= nb_clusters
* s
->cluster_size
;
3840 host_offset
+= nb_clusters
* s
->cluster_size
;
3841 nb_new_data_clusters
-= nb_clusters
;
3847 g_assert_not_reached();
3850 if (prealloc
!= PREALLOC_MODE_OFF
) {
3851 /* Flush metadata before actually changing the image size */
3852 ret
= qcow2_write_caches(bs
);
3854 error_setg_errno(errp
, -ret
,
3855 "Failed to flush the preallocated area to disk");
3860 bs
->total_sectors
= offset
/ BDRV_SECTOR_SIZE
;
3862 if (has_data_file(bs
)) {
3863 if (prealloc
== PREALLOC_MODE_METADATA
) {
3864 prealloc
= PREALLOC_MODE_OFF
;
3866 ret
= bdrv_co_truncate(s
->data_file
, offset
, prealloc
, errp
);
3872 /* write updated header.size */
3873 offset
= cpu_to_be64(offset
);
3874 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, size
),
3875 &offset
, sizeof(uint64_t));
3877 error_setg_errno(errp
, -ret
, "Failed to update the image size");
3881 s
->l1_vm_state_index
= new_l1_size
;
3883 /* Update cache sizes */
3884 options
= qdict_clone_shallow(bs
->options
);
3885 ret
= qcow2_update_options(bs
, options
, s
->flags
, errp
);
3886 qobject_unref(options
);
3892 qemu_co_mutex_unlock(&s
->lock
);
3899 * @dest - destination buffer, @dest_size bytes
3900 * @src - source buffer, @src_size bytes
3902 * Returns: compressed size on success
3903 * -1 destination buffer is not enough to store compressed data
3904 * -2 on any other error
3906 static ssize_t
qcow2_compress(void *dest
, size_t dest_size
,
3907 const void *src
, size_t src_size
)
3912 /* best compression, small window, no zlib header */
3913 memset(&strm
, 0, sizeof(strm
));
3914 ret
= deflateInit2(&strm
, Z_DEFAULT_COMPRESSION
, Z_DEFLATED
,
3915 -12, 9, Z_DEFAULT_STRATEGY
);
3920 /* strm.next_in is not const in old zlib versions, such as those used on
3921 * OpenBSD/NetBSD, so cast the const away */
3922 strm
.avail_in
= src_size
;
3923 strm
.next_in
= (void *) src
;
3924 strm
.avail_out
= dest_size
;
3925 strm
.next_out
= dest
;
3927 ret
= deflate(&strm
, Z_FINISH
);
3928 if (ret
== Z_STREAM_END
) {
3929 ret
= dest_size
- strm
.avail_out
;
3931 ret
= (ret
== Z_OK
? -1 : -2);
3940 * qcow2_decompress()
3942 * Decompress some data (not more than @src_size bytes) to produce exactly
3945 * @dest - destination buffer, @dest_size bytes
3946 * @src - source buffer, @src_size bytes
3948 * Returns: 0 on success
3951 static ssize_t
qcow2_decompress(void *dest
, size_t dest_size
,
3952 const void *src
, size_t src_size
)
3957 memset(&strm
, 0, sizeof(strm
));
3958 strm
.avail_in
= src_size
;
3959 strm
.next_in
= (void *) src
;
3960 strm
.avail_out
= dest_size
;
3961 strm
.next_out
= dest
;
3963 ret
= inflateInit2(&strm
, -12);
3968 ret
= inflate(&strm
, Z_FINISH
);
3969 if ((ret
!= Z_STREAM_END
&& ret
!= Z_BUF_ERROR
) || strm
.avail_out
!= 0) {
3970 /* We approve Z_BUF_ERROR because we need @dest buffer to be filled, but
3971 * @src buffer may be processed partly (because in qcow2 we know size of
3972 * compressed data with precision of one sector) */
3981 #define MAX_COMPRESS_THREADS 4
3983 typedef ssize_t (*Qcow2CompressFunc
)(void *dest
, size_t dest_size
,
3984 const void *src
, size_t src_size
);
3985 typedef struct Qcow2CompressData
{
3992 Qcow2CompressFunc func
;
3993 } Qcow2CompressData
;
3995 static int qcow2_compress_pool_func(void *opaque
)
3997 Qcow2CompressData
*data
= opaque
;
3999 data
->ret
= data
->func(data
->dest
, data
->dest_size
,
4000 data
->src
, data
->src_size
);
4005 static void qcow2_compress_complete(void *opaque
, int ret
)
4007 qemu_coroutine_enter(opaque
);
4010 static ssize_t coroutine_fn
4011 qcow2_co_do_compress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
4012 const void *src
, size_t src_size
, Qcow2CompressFunc func
)
4014 BDRVQcow2State
*s
= bs
->opaque
;
4016 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
4017 Qcow2CompressData arg
= {
4019 .dest_size
= dest_size
,
4021 .src_size
= src_size
,
4025 while (s
->nb_compress_threads
>= MAX_COMPRESS_THREADS
) {
4026 qemu_co_queue_wait(&s
->compress_wait_queue
, NULL
);
4029 s
->nb_compress_threads
++;
4030 acb
= thread_pool_submit_aio(pool
, qcow2_compress_pool_func
, &arg
,
4031 qcow2_compress_complete
,
4032 qemu_coroutine_self());
4035 s
->nb_compress_threads
--;
4038 qemu_coroutine_yield();
4039 s
->nb_compress_threads
--;
4040 qemu_co_queue_next(&s
->compress_wait_queue
);
4045 static ssize_t coroutine_fn
4046 qcow2_co_compress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
4047 const void *src
, size_t src_size
)
4049 return qcow2_co_do_compress(bs
, dest
, dest_size
, src
, src_size
,
4053 static ssize_t coroutine_fn
4054 qcow2_co_decompress(BlockDriverState
*bs
, void *dest
, size_t dest_size
,
4055 const void *src
, size_t src_size
)
4057 return qcow2_co_do_compress(bs
, dest
, dest_size
, src
, src_size
,
4061 /* XXX: put compressed sectors first, then all the cluster aligned
4062 tables to avoid losing bytes in alignment */
4063 static coroutine_fn
int
4064 qcow2_co_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
4065 uint64_t bytes
, QEMUIOVector
*qiov
)
4067 BDRVQcow2State
*s
= bs
->opaque
;
4068 QEMUIOVector hd_qiov
;
4071 uint8_t *buf
, *out_buf
;
4072 uint64_t cluster_offset
;
4074 if (has_data_file(bs
)) {
4079 /* align end of file to a sector boundary to ease reading with
4080 sector based I/Os */
4081 int64_t len
= bdrv_getlength(bs
->file
->bs
);
4085 return bdrv_co_truncate(bs
->file
, len
, PREALLOC_MODE_OFF
, NULL
);
4088 if (offset_into_cluster(s
, offset
)) {
4092 buf
= qemu_blockalign(bs
, s
->cluster_size
);
4093 if (bytes
!= s
->cluster_size
) {
4094 if (bytes
> s
->cluster_size
||
4095 offset
+ bytes
!= bs
->total_sectors
<< BDRV_SECTOR_BITS
)
4100 /* Zero-pad last write if image size is not cluster aligned */
4101 memset(buf
+ bytes
, 0, s
->cluster_size
- bytes
);
4103 qemu_iovec_to_buf(qiov
, 0, buf
, bytes
);
4105 out_buf
= g_malloc(s
->cluster_size
);
4107 out_len
= qcow2_co_compress(bs
, out_buf
, s
->cluster_size
- 1,
4108 buf
, s
->cluster_size
);
4109 if (out_len
== -2) {
4112 } else if (out_len
== -1) {
4113 /* could not compress: write normal cluster */
4114 ret
= qcow2_co_pwritev(bs
, offset
, bytes
, qiov
, 0);
4121 qemu_co_mutex_lock(&s
->lock
);
4122 ret
= qcow2_alloc_compressed_cluster_offset(bs
, offset
, out_len
,
4125 qemu_co_mutex_unlock(&s
->lock
);
4129 ret
= qcow2_pre_write_overlap_check(bs
, 0, cluster_offset
, out_len
, true);
4130 qemu_co_mutex_unlock(&s
->lock
);
4135 qemu_iovec_init_buf(&hd_qiov
, out_buf
, out_len
);
4137 BLKDBG_EVENT(s
->data_file
, BLKDBG_WRITE_COMPRESSED
);
4138 ret
= bdrv_co_pwritev(s
->data_file
, cluster_offset
, out_len
, &hd_qiov
, 0);
4150 static int coroutine_fn
4151 qcow2_co_preadv_compressed(BlockDriverState
*bs
,
4152 uint64_t file_cluster_offset
,
4157 BDRVQcow2State
*s
= bs
->opaque
;
4158 int ret
= 0, csize
, nb_csectors
;
4160 uint8_t *buf
, *out_buf
;
4161 QEMUIOVector local_qiov
;
4162 int offset_in_cluster
= offset_into_cluster(s
, offset
);
4164 coffset
= file_cluster_offset
& s
->cluster_offset_mask
;
4165 nb_csectors
= ((file_cluster_offset
>> s
->csize_shift
) & s
->csize_mask
) + 1;
4166 csize
= nb_csectors
* 512 - (coffset
& 511);
4168 buf
= g_try_malloc(csize
);
4172 qemu_iovec_init_buf(&local_qiov
, buf
, csize
);
4174 out_buf
= qemu_blockalign(bs
, s
->cluster_size
);
4176 BLKDBG_EVENT(bs
->file
, BLKDBG_READ_COMPRESSED
);
4177 ret
= bdrv_co_preadv(bs
->file
, coffset
, csize
, &local_qiov
, 0);
4182 if (qcow2_co_decompress(bs
, out_buf
, s
->cluster_size
, buf
, csize
) < 0) {
4187 qemu_iovec_from_buf(qiov
, 0, out_buf
+ offset_in_cluster
, bytes
);
4190 qemu_vfree(out_buf
);
4196 static int make_completely_empty(BlockDriverState
*bs
)
4198 BDRVQcow2State
*s
= bs
->opaque
;
4199 Error
*local_err
= NULL
;
4200 int ret
, l1_clusters
;
4202 uint64_t *new_reftable
= NULL
;
4203 uint64_t rt_entry
, l1_size2
;
4206 uint64_t reftable_offset
;
4207 uint32_t reftable_clusters
;
4208 } QEMU_PACKED l1_ofs_rt_ofs_cls
;
4210 ret
= qcow2_cache_empty(bs
, s
->l2_table_cache
);
4215 ret
= qcow2_cache_empty(bs
, s
->refcount_block_cache
);
4220 /* Refcounts will be broken utterly */
4221 ret
= qcow2_mark_dirty(bs
);
4226 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4228 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4229 l1_size2
= (uint64_t)s
->l1_size
* sizeof(uint64_t);
4231 /* After this call, neither the in-memory nor the on-disk refcount
4232 * information accurately describe the actual references */
4234 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->l1_table_offset
,
4235 l1_clusters
* s
->cluster_size
, 0);
4237 goto fail_broken_refcounts
;
4239 memset(s
->l1_table
, 0, l1_size2
);
4241 BLKDBG_EVENT(bs
->file
, BLKDBG_EMPTY_IMAGE_PREPARE
);
4243 /* Overwrite enough clusters at the beginning of the sectors to place
4244 * the refcount table, a refcount block and the L1 table in; this may
4245 * overwrite parts of the existing refcount and L1 table, which is not
4246 * an issue because the dirty flag is set, complete data loss is in fact
4247 * desired and partial data loss is consequently fine as well */
4248 ret
= bdrv_pwrite_zeroes(bs
->file
, s
->cluster_size
,
4249 (2 + l1_clusters
) * s
->cluster_size
, 0);
4250 /* This call (even if it failed overall) may have overwritten on-disk
4251 * refcount structures; in that case, the in-memory refcount information
4252 * will probably differ from the on-disk information which makes the BDS
4255 goto fail_broken_refcounts
;
4258 BLKDBG_EVENT(bs
->file
, BLKDBG_L1_UPDATE
);
4259 BLKDBG_EVENT(bs
->file
, BLKDBG_REFTABLE_UPDATE
);
4261 /* "Create" an empty reftable (one cluster) directly after the image
4262 * header and an empty L1 table three clusters after the image header;
4263 * the cluster between those two will be used as the first refblock */
4264 l1_ofs_rt_ofs_cls
.l1_offset
= cpu_to_be64(3 * s
->cluster_size
);
4265 l1_ofs_rt_ofs_cls
.reftable_offset
= cpu_to_be64(s
->cluster_size
);
4266 l1_ofs_rt_ofs_cls
.reftable_clusters
= cpu_to_be32(1);
4267 ret
= bdrv_pwrite_sync(bs
->file
, offsetof(QCowHeader
, l1_table_offset
),
4268 &l1_ofs_rt_ofs_cls
, sizeof(l1_ofs_rt_ofs_cls
));
4270 goto fail_broken_refcounts
;
4273 s
->l1_table_offset
= 3 * s
->cluster_size
;
4275 new_reftable
= g_try_new0(uint64_t, s
->cluster_size
/ sizeof(uint64_t));
4276 if (!new_reftable
) {
4278 goto fail_broken_refcounts
;
4281 s
->refcount_table_offset
= s
->cluster_size
;
4282 s
->refcount_table_size
= s
->cluster_size
/ sizeof(uint64_t);
4283 s
->max_refcount_table_index
= 0;
4285 g_free(s
->refcount_table
);
4286 s
->refcount_table
= new_reftable
;
4287 new_reftable
= NULL
;
4289 /* Now the in-memory refcount information again corresponds to the on-disk
4290 * information (reftable is empty and no refblocks (the refblock cache is
4291 * empty)); however, this means some clusters (e.g. the image header) are
4292 * referenced, but not refcounted, but the normal qcow2 code assumes that
4293 * the in-memory information is always correct */
4295 BLKDBG_EVENT(bs
->file
, BLKDBG_REFBLOCK_ALLOC
);
4297 /* Enter the first refblock into the reftable */
4298 rt_entry
= cpu_to_be64(2 * s
->cluster_size
);
4299 ret
= bdrv_pwrite_sync(bs
->file
, s
->cluster_size
,
4300 &rt_entry
, sizeof(rt_entry
));
4302 goto fail_broken_refcounts
;
4304 s
->refcount_table
[0] = 2 * s
->cluster_size
;
4306 s
->free_cluster_index
= 0;
4307 assert(3 + l1_clusters
<= s
->refcount_block_size
);
4308 offset
= qcow2_alloc_clusters(bs
, 3 * s
->cluster_size
+ l1_size2
);
4311 goto fail_broken_refcounts
;
4312 } else if (offset
> 0) {
4313 error_report("First cluster in emptied image is in use");
4317 /* Now finally the in-memory information corresponds to the on-disk
4318 * structures and is correct */
4319 ret
= qcow2_mark_clean(bs
);
4324 ret
= bdrv_truncate(bs
->file
, (3 + l1_clusters
) * s
->cluster_size
,
4325 PREALLOC_MODE_OFF
, &local_err
);
4327 error_report_err(local_err
);
4333 fail_broken_refcounts
:
4334 /* The BDS is unusable at this point. If we wanted to make it usable, we
4335 * would have to call qcow2_refcount_close(), qcow2_refcount_init(),
4336 * qcow2_check_refcounts(), qcow2_refcount_close() and qcow2_refcount_init()
4337 * again. However, because the functions which could have caused this error
4338 * path to be taken are used by those functions as well, it's very likely
4339 * that that sequence will fail as well. Therefore, just eject the BDS. */
4343 g_free(new_reftable
);
4347 static int qcow2_make_empty(BlockDriverState
*bs
)
4349 BDRVQcow2State
*s
= bs
->opaque
;
4350 uint64_t offset
, end_offset
;
4351 int step
= QEMU_ALIGN_DOWN(INT_MAX
, s
->cluster_size
);
4352 int l1_clusters
, ret
= 0;
4354 l1_clusters
= DIV_ROUND_UP(s
->l1_size
, s
->cluster_size
/ sizeof(uint64_t));
4356 if (s
->qcow_version
>= 3 && !s
->snapshots
&& !s
->nb_bitmaps
&&
4357 3 + l1_clusters
<= s
->refcount_block_size
&&
4358 s
->crypt_method_header
!= QCOW_CRYPT_LUKS
) {
4359 /* The following function only works for qcow2 v3 images (it
4360 * requires the dirty flag) and only as long as there are no
4361 * features that reserve extra clusters (such as snapshots,
4362 * LUKS header, or persistent bitmaps), because it completely
4363 * empties the image. Furthermore, the L1 table and three
4364 * additional clusters (image header, refcount table, one
4365 * refcount block) have to fit inside one refcount block. */
4366 return make_completely_empty(bs
);
4369 /* This fallback code simply discards every active cluster; this is slow,
4370 * but works in all cases */
4371 end_offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
4372 for (offset
= 0; offset
< end_offset
; offset
+= step
) {
4373 /* As this function is generally used after committing an external
4374 * snapshot, QCOW2_DISCARD_SNAPSHOT seems appropriate. Also, the
4375 * default action for this kind of discard is to pass the discard,
4376 * which will ideally result in an actually smaller image file, as
4377 * is probably desired. */
4378 ret
= qcow2_cluster_discard(bs
, offset
, MIN(step
, end_offset
- offset
),
4379 QCOW2_DISCARD_SNAPSHOT
, true);
4388 static coroutine_fn
int qcow2_co_flush_to_os(BlockDriverState
*bs
)
4390 BDRVQcow2State
*s
= bs
->opaque
;
4393 qemu_co_mutex_lock(&s
->lock
);
4394 ret
= qcow2_write_caches(bs
);
4395 qemu_co_mutex_unlock(&s
->lock
);
4400 static ssize_t
qcow2_measure_crypto_hdr_init_func(QCryptoBlock
*block
,
4401 size_t headerlen
, void *opaque
, Error
**errp
)
4403 size_t *headerlenp
= opaque
;
4405 /* Stash away the payload size */
4406 *headerlenp
= headerlen
;
4410 static ssize_t
qcow2_measure_crypto_hdr_write_func(QCryptoBlock
*block
,
4411 size_t offset
, const uint8_t *buf
, size_t buflen
,
4412 void *opaque
, Error
**errp
)
4414 /* Discard the bytes, we're not actually writing to an image */
4418 /* Determine the number of bytes for the LUKS payload */
4419 static bool qcow2_measure_luks_headerlen(QemuOpts
*opts
, size_t *len
,
4423 QDict
*cryptoopts_qdict
;
4424 QCryptoBlockCreateOptions
*cryptoopts
;
4425 QCryptoBlock
*crypto
;
4427 /* Extract "encrypt." options into a qdict */
4428 opts_qdict
= qemu_opts_to_qdict(opts
, NULL
);
4429 qdict_extract_subqdict(opts_qdict
, &cryptoopts_qdict
, "encrypt.");
4430 qobject_unref(opts_qdict
);
4432 /* Build QCryptoBlockCreateOptions object from qdict */
4433 qdict_put_str(cryptoopts_qdict
, "format", "luks");
4434 cryptoopts
= block_crypto_create_opts_init(cryptoopts_qdict
, errp
);
4435 qobject_unref(cryptoopts_qdict
);
4440 /* Fake LUKS creation in order to determine the payload size */
4441 crypto
= qcrypto_block_create(cryptoopts
, "encrypt.",
4442 qcow2_measure_crypto_hdr_init_func
,
4443 qcow2_measure_crypto_hdr_write_func
,
4445 qapi_free_QCryptoBlockCreateOptions(cryptoopts
);
4450 qcrypto_block_free(crypto
);
4454 static BlockMeasureInfo
*qcow2_measure(QemuOpts
*opts
, BlockDriverState
*in_bs
,
4457 Error
*local_err
= NULL
;
4458 BlockMeasureInfo
*info
;
4459 uint64_t required
= 0; /* bytes that contribute to required size */
4460 uint64_t virtual_size
; /* disk size as seen by guest */
4461 uint64_t refcount_bits
;
4463 uint64_t luks_payload_size
= 0;
4464 size_t cluster_size
;
4467 PreallocMode prealloc
;
4468 bool has_backing_file
;
4471 /* Parse image creation options */
4472 cluster_size
= qcow2_opt_get_cluster_size_del(opts
, &local_err
);
4477 version
= qcow2_opt_get_version_del(opts
, &local_err
);
4482 refcount_bits
= qcow2_opt_get_refcount_bits_del(opts
, version
, &local_err
);
4487 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
4488 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, optstr
,
4489 PREALLOC_MODE_OFF
, &local_err
);
4495 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_BACKING_FILE
);
4496 has_backing_file
= !!optstr
;
4499 optstr
= qemu_opt_get_del(opts
, BLOCK_OPT_ENCRYPT_FORMAT
);
4500 has_luks
= optstr
&& strcmp(optstr
, "luks") == 0;
4506 if (!qcow2_measure_luks_headerlen(opts
, &headerlen
, &local_err
)) {
4510 luks_payload_size
= ROUND_UP(headerlen
, cluster_size
);
4513 virtual_size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
4514 virtual_size
= ROUND_UP(virtual_size
, cluster_size
);
4516 /* Check that virtual disk size is valid */
4517 l2_tables
= DIV_ROUND_UP(virtual_size
/ cluster_size
,
4518 cluster_size
/ sizeof(uint64_t));
4519 if (l2_tables
* sizeof(uint64_t) > QCOW_MAX_L1_SIZE
) {
4520 error_setg(&local_err
, "The image size is too large "
4521 "(try using a larger cluster size)");
4525 /* Account for input image */
4527 int64_t ssize
= bdrv_getlength(in_bs
);
4529 error_setg_errno(&local_err
, -ssize
,
4530 "Unable to get image virtual_size");
4534 virtual_size
= ROUND_UP(ssize
, cluster_size
);
4536 if (has_backing_file
) {
4537 /* We don't how much of the backing chain is shared by the input
4538 * image and the new image file. In the worst case the new image's
4539 * backing file has nothing in common with the input image. Be
4540 * conservative and assume all clusters need to be written.
4542 required
= virtual_size
;
4547 for (offset
= 0; offset
< ssize
; offset
+= pnum
) {
4550 ret
= bdrv_block_status_above(in_bs
, NULL
, offset
,
4551 ssize
- offset
, &pnum
, NULL
,
4554 error_setg_errno(&local_err
, -ret
,
4555 "Unable to get block status");
4559 if (ret
& BDRV_BLOCK_ZERO
) {
4560 /* Skip zero regions (safe with no backing file) */
4561 } else if ((ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) ==
4562 (BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
)) {
4563 /* Extend pnum to end of cluster for next iteration */
4564 pnum
= ROUND_UP(offset
+ pnum
, cluster_size
) - offset
;
4566 /* Count clusters we've seen */
4567 required
+= offset
% cluster_size
+ pnum
;
4573 /* Take into account preallocation. Nothing special is needed for
4574 * PREALLOC_MODE_METADATA since metadata is always counted.
4576 if (prealloc
== PREALLOC_MODE_FULL
|| prealloc
== PREALLOC_MODE_FALLOC
) {
4577 required
= virtual_size
;
4580 info
= g_new(BlockMeasureInfo
, 1);
4581 info
->fully_allocated
=
4582 qcow2_calc_prealloc_size(virtual_size
, cluster_size
,
4583 ctz32(refcount_bits
)) + luks_payload_size
;
4585 /* Remove data clusters that are not required. This overestimates the
4586 * required size because metadata needed for the fully allocated file is
4589 info
->required
= info
->fully_allocated
- virtual_size
+ required
;
4593 error_propagate(errp
, local_err
);
4597 static int qcow2_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4599 BDRVQcow2State
*s
= bs
->opaque
;
4600 bdi
->unallocated_blocks_are_zero
= true;
4601 bdi
->cluster_size
= s
->cluster_size
;
4602 bdi
->vm_state_offset
= qcow2_vm_state_offset(s
);
4606 static ImageInfoSpecific
*qcow2_get_specific_info(BlockDriverState
*bs
,
4609 BDRVQcow2State
*s
= bs
->opaque
;
4610 ImageInfoSpecific
*spec_info
;
4611 QCryptoBlockInfo
*encrypt_info
= NULL
;
4612 Error
*local_err
= NULL
;
4614 if (s
->crypto
!= NULL
) {
4615 encrypt_info
= qcrypto_block_get_info(s
->crypto
, &local_err
);
4617 error_propagate(errp
, local_err
);
4622 spec_info
= g_new(ImageInfoSpecific
, 1);
4623 *spec_info
= (ImageInfoSpecific
){
4624 .type
= IMAGE_INFO_SPECIFIC_KIND_QCOW2
,
4625 .u
.qcow2
.data
= g_new0(ImageInfoSpecificQCow2
, 1),
4627 if (s
->qcow_version
== 2) {
4628 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4629 .compat
= g_strdup("0.10"),
4630 .refcount_bits
= s
->refcount_bits
,
4632 } else if (s
->qcow_version
== 3) {
4633 Qcow2BitmapInfoList
*bitmaps
;
4634 bitmaps
= qcow2_get_bitmap_info_list(bs
, &local_err
);
4636 error_propagate(errp
, local_err
);
4637 qapi_free_ImageInfoSpecific(spec_info
);
4640 *spec_info
->u
.qcow2
.data
= (ImageInfoSpecificQCow2
){
4641 .compat
= g_strdup("1.1"),
4642 .lazy_refcounts
= s
->compatible_features
&
4643 QCOW2_COMPAT_LAZY_REFCOUNTS
,
4644 .has_lazy_refcounts
= true,
4645 .corrupt
= s
->incompatible_features
&
4646 QCOW2_INCOMPAT_CORRUPT
,
4647 .has_corrupt
= true,
4648 .refcount_bits
= s
->refcount_bits
,
4649 .has_bitmaps
= !!bitmaps
,
4651 .has_data_file
= !!s
->image_data_file
,
4652 .data_file
= g_strdup(s
->image_data_file
),
4653 .has_data_file_raw
= has_data_file(bs
),
4654 .data_file_raw
= data_file_is_raw(bs
),
4657 /* if this assertion fails, this probably means a new version was
4658 * added without having it covered here */
4663 ImageInfoSpecificQCow2Encryption
*qencrypt
=
4664 g_new(ImageInfoSpecificQCow2Encryption
, 1);
4665 switch (encrypt_info
->format
) {
4666 case Q_CRYPTO_BLOCK_FORMAT_QCOW
:
4667 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_AES
;
4669 case Q_CRYPTO_BLOCK_FORMAT_LUKS
:
4670 qencrypt
->format
= BLOCKDEV_QCOW2_ENCRYPTION_FORMAT_LUKS
;
4671 qencrypt
->u
.luks
= encrypt_info
->u
.luks
;
4676 /* Since we did shallow copy above, erase any pointers
4677 * in the original info */
4678 memset(&encrypt_info
->u
, 0, sizeof(encrypt_info
->u
));
4679 qapi_free_QCryptoBlockInfo(encrypt_info
);
4681 spec_info
->u
.qcow2
.data
->has_encrypt
= true;
4682 spec_info
->u
.qcow2
.data
->encrypt
= qencrypt
;
4688 static int qcow2_save_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4691 BDRVQcow2State
*s
= bs
->opaque
;
4693 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_SAVE
);
4694 return bs
->drv
->bdrv_co_pwritev(bs
, qcow2_vm_state_offset(s
) + pos
,
4695 qiov
->size
, qiov
, 0);
4698 static int qcow2_load_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
,
4701 BDRVQcow2State
*s
= bs
->opaque
;
4703 BLKDBG_EVENT(bs
->file
, BLKDBG_VMSTATE_LOAD
);
4704 return bs
->drv
->bdrv_co_preadv(bs
, qcow2_vm_state_offset(s
) + pos
,
4705 qiov
->size
, qiov
, 0);
4709 * Downgrades an image's version. To achieve this, any incompatible features
4710 * have to be removed.
4712 static int qcow2_downgrade(BlockDriverState
*bs
, int target_version
,
4713 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
4716 BDRVQcow2State
*s
= bs
->opaque
;
4717 int current_version
= s
->qcow_version
;
4720 /* This is qcow2_downgrade(), not qcow2_upgrade() */
4721 assert(target_version
< current_version
);
4723 /* There are no other versions (now) that you can downgrade to */
4724 assert(target_version
== 2);
4726 if (s
->refcount_order
!= 4) {
4727 error_setg(errp
, "compat=0.10 requires refcount_bits=16");
4731 if (has_data_file(bs
)) {
4732 error_setg(errp
, "Cannot downgrade an image with a data file");
4736 /* clear incompatible features */
4737 if (s
->incompatible_features
& QCOW2_INCOMPAT_DIRTY
) {
4738 ret
= qcow2_mark_clean(bs
);
4740 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
4745 /* with QCOW2_INCOMPAT_CORRUPT, it is pretty much impossible to get here in
4746 * the first place; if that happens nonetheless, returning -ENOTSUP is the
4747 * best thing to do anyway */
4749 if (s
->incompatible_features
) {
4750 error_setg(errp
, "Cannot downgrade an image with incompatible features "
4751 "%#" PRIx64
" set", s
->incompatible_features
);
4755 /* since we can ignore compatible features, we can set them to 0 as well */
4756 s
->compatible_features
= 0;
4757 /* if lazy refcounts have been used, they have already been fixed through
4758 * clearing the dirty flag */
4760 /* clearing autoclear features is trivial */
4761 s
->autoclear_features
= 0;
4763 ret
= qcow2_expand_zero_clusters(bs
, status_cb
, cb_opaque
);
4765 error_setg_errno(errp
, -ret
, "Failed to turn zero into data clusters");
4769 s
->qcow_version
= target_version
;
4770 ret
= qcow2_update_header(bs
);
4772 s
->qcow_version
= current_version
;
4773 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4779 typedef enum Qcow2AmendOperation
{
4780 /* This is the value Qcow2AmendHelperCBInfo::last_operation will be
4781 * statically initialized to so that the helper CB can discern the first
4782 * invocation from an operation change */
4783 QCOW2_NO_OPERATION
= 0,
4785 QCOW2_CHANGING_REFCOUNT_ORDER
,
4787 } Qcow2AmendOperation
;
4789 typedef struct Qcow2AmendHelperCBInfo
{
4790 /* The code coordinating the amend operations should only modify
4791 * these four fields; the rest will be managed by the CB */
4792 BlockDriverAmendStatusCB
*original_status_cb
;
4793 void *original_cb_opaque
;
4795 Qcow2AmendOperation current_operation
;
4797 /* Total number of operations to perform (only set once) */
4798 int total_operations
;
4800 /* The following fields are managed by the CB */
4802 /* Number of operations completed */
4803 int operations_completed
;
4805 /* Cumulative offset of all completed operations */
4806 int64_t offset_completed
;
4808 Qcow2AmendOperation last_operation
;
4809 int64_t last_work_size
;
4810 } Qcow2AmendHelperCBInfo
;
4812 static void qcow2_amend_helper_cb(BlockDriverState
*bs
,
4813 int64_t operation_offset
,
4814 int64_t operation_work_size
, void *opaque
)
4816 Qcow2AmendHelperCBInfo
*info
= opaque
;
4817 int64_t current_work_size
;
4818 int64_t projected_work_size
;
4820 if (info
->current_operation
!= info
->last_operation
) {
4821 if (info
->last_operation
!= QCOW2_NO_OPERATION
) {
4822 info
->offset_completed
+= info
->last_work_size
;
4823 info
->operations_completed
++;
4826 info
->last_operation
= info
->current_operation
;
4829 assert(info
->total_operations
> 0);
4830 assert(info
->operations_completed
< info
->total_operations
);
4832 info
->last_work_size
= operation_work_size
;
4834 current_work_size
= info
->offset_completed
+ operation_work_size
;
4836 /* current_work_size is the total work size for (operations_completed + 1)
4837 * operations (which includes this one), so multiply it by the number of
4838 * operations not covered and divide it by the number of operations
4839 * covered to get a projection for the operations not covered */
4840 projected_work_size
= current_work_size
* (info
->total_operations
-
4841 info
->operations_completed
- 1)
4842 / (info
->operations_completed
+ 1);
4844 info
->original_status_cb(bs
, info
->offset_completed
+ operation_offset
,
4845 current_work_size
+ projected_work_size
,
4846 info
->original_cb_opaque
);
4849 static int qcow2_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4850 BlockDriverAmendStatusCB
*status_cb
,
4854 BDRVQcow2State
*s
= bs
->opaque
;
4855 int old_version
= s
->qcow_version
, new_version
= old_version
;
4856 uint64_t new_size
= 0;
4857 const char *backing_file
= NULL
, *backing_format
= NULL
, *data_file
= NULL
;
4858 bool lazy_refcounts
= s
->use_lazy_refcounts
;
4859 bool data_file_raw
= data_file_is_raw(bs
);
4860 const char *compat
= NULL
;
4861 uint64_t cluster_size
= s
->cluster_size
;
4864 int refcount_bits
= s
->refcount_bits
;
4866 QemuOptDesc
*desc
= opts
->list
->desc
;
4867 Qcow2AmendHelperCBInfo helper_cb_info
;
4869 while (desc
&& desc
->name
) {
4870 if (!qemu_opt_find(opts
, desc
->name
)) {
4871 /* only change explicitly defined options */
4876 if (!strcmp(desc
->name
, BLOCK_OPT_COMPAT_LEVEL
)) {
4877 compat
= qemu_opt_get(opts
, BLOCK_OPT_COMPAT_LEVEL
);
4879 /* preserve default */
4880 } else if (!strcmp(compat
, "0.10")) {
4882 } else if (!strcmp(compat
, "1.1")) {
4885 error_setg(errp
, "Unknown compatibility level %s", compat
);
4888 } else if (!strcmp(desc
->name
, BLOCK_OPT_PREALLOC
)) {
4889 error_setg(errp
, "Cannot change preallocation mode");
4891 } else if (!strcmp(desc
->name
, BLOCK_OPT_SIZE
)) {
4892 new_size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4893 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FILE
)) {
4894 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4895 } else if (!strcmp(desc
->name
, BLOCK_OPT_BACKING_FMT
)) {
4896 backing_format
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4897 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT
)) {
4898 encrypt
= qemu_opt_get_bool(opts
, BLOCK_OPT_ENCRYPT
,
4901 if (encrypt
!= !!s
->crypto
) {
4903 "Changing the encryption flag is not supported");
4906 } else if (!strcmp(desc
->name
, BLOCK_OPT_ENCRYPT_FORMAT
)) {
4907 encformat
= qcow2_crypt_method_from_format(
4908 qemu_opt_get(opts
, BLOCK_OPT_ENCRYPT_FORMAT
));
4910 if (encformat
!= s
->crypt_method_header
) {
4912 "Changing the encryption format is not supported");
4915 } else if (g_str_has_prefix(desc
->name
, "encrypt.")) {
4917 "Changing the encryption parameters is not supported");
4919 } else if (!strcmp(desc
->name
, BLOCK_OPT_CLUSTER_SIZE
)) {
4920 cluster_size
= qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
,
4922 if (cluster_size
!= s
->cluster_size
) {
4923 error_setg(errp
, "Changing the cluster size is not supported");
4926 } else if (!strcmp(desc
->name
, BLOCK_OPT_LAZY_REFCOUNTS
)) {
4927 lazy_refcounts
= qemu_opt_get_bool(opts
, BLOCK_OPT_LAZY_REFCOUNTS
,
4929 } else if (!strcmp(desc
->name
, BLOCK_OPT_REFCOUNT_BITS
)) {
4930 refcount_bits
= qemu_opt_get_number(opts
, BLOCK_OPT_REFCOUNT_BITS
,
4933 if (refcount_bits
<= 0 || refcount_bits
> 64 ||
4934 !is_power_of_2(refcount_bits
))
4936 error_setg(errp
, "Refcount width must be a power of two and "
4937 "may not exceed 64 bits");
4940 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE
)) {
4941 data_file
= qemu_opt_get(opts
, BLOCK_OPT_DATA_FILE
);
4942 if (data_file
&& !has_data_file(bs
)) {
4943 error_setg(errp
, "data-file can only be set for images that "
4944 "use an external data file");
4947 } else if (!strcmp(desc
->name
, BLOCK_OPT_DATA_FILE_RAW
)) {
4948 data_file_raw
= qemu_opt_get_bool(opts
, BLOCK_OPT_DATA_FILE_RAW
,
4950 if (data_file_raw
&& !data_file_is_raw(bs
)) {
4951 error_setg(errp
, "data-file-raw cannot be set on existing "
4956 /* if this point is reached, this probably means a new option was
4957 * added without having it covered here */
4964 helper_cb_info
= (Qcow2AmendHelperCBInfo
){
4965 .original_status_cb
= status_cb
,
4966 .original_cb_opaque
= cb_opaque
,
4967 .total_operations
= (new_version
< old_version
)
4968 + (s
->refcount_bits
!= refcount_bits
)
4971 /* Upgrade first (some features may require compat=1.1) */
4972 if (new_version
> old_version
) {
4973 s
->qcow_version
= new_version
;
4974 ret
= qcow2_update_header(bs
);
4976 s
->qcow_version
= old_version
;
4977 error_setg_errno(errp
, -ret
, "Failed to update the image header");
4982 if (s
->refcount_bits
!= refcount_bits
) {
4983 int refcount_order
= ctz32(refcount_bits
);
4985 if (new_version
< 3 && refcount_bits
!= 16) {
4986 error_setg(errp
, "Refcount widths other than 16 bits require "
4987 "compatibility level 1.1 or above (use compat=1.1 or "
4992 helper_cb_info
.current_operation
= QCOW2_CHANGING_REFCOUNT_ORDER
;
4993 ret
= qcow2_change_refcount_order(bs
, refcount_order
,
4994 &qcow2_amend_helper_cb
,
4995 &helper_cb_info
, errp
);
5001 /* data-file-raw blocks backing files, so clear it first if requested */
5002 if (data_file_raw
) {
5003 s
->autoclear_features
|= QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5005 s
->autoclear_features
&= ~QCOW2_AUTOCLEAR_DATA_FILE_RAW
;
5009 g_free(s
->image_data_file
);
5010 s
->image_data_file
= *data_file
? g_strdup(data_file
) : NULL
;
5013 ret
= qcow2_update_header(bs
);
5015 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5019 if (backing_file
|| backing_format
) {
5020 ret
= qcow2_change_backing_file(bs
,
5021 backing_file
?: s
->image_backing_file
,
5022 backing_format
?: s
->image_backing_format
);
5024 error_setg_errno(errp
, -ret
, "Failed to change the backing file");
5029 if (s
->use_lazy_refcounts
!= lazy_refcounts
) {
5030 if (lazy_refcounts
) {
5031 if (new_version
< 3) {
5032 error_setg(errp
, "Lazy refcounts only supported with "
5033 "compatibility level 1.1 and above (use compat=1.1 "
5037 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5038 ret
= qcow2_update_header(bs
);
5040 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5041 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5044 s
->use_lazy_refcounts
= true;
5046 /* make image clean first */
5047 ret
= qcow2_mark_clean(bs
);
5049 error_setg_errno(errp
, -ret
, "Failed to make the image clean");
5052 /* now disallow lazy refcounts */
5053 s
->compatible_features
&= ~QCOW2_COMPAT_LAZY_REFCOUNTS
;
5054 ret
= qcow2_update_header(bs
);
5056 s
->compatible_features
|= QCOW2_COMPAT_LAZY_REFCOUNTS
;
5057 error_setg_errno(errp
, -ret
, "Failed to update the image header");
5060 s
->use_lazy_refcounts
= false;
5065 BlockBackend
*blk
= blk_new(BLK_PERM_RESIZE
, BLK_PERM_ALL
);
5066 ret
= blk_insert_bs(blk
, bs
, errp
);
5072 ret
= blk_truncate(blk
, new_size
, PREALLOC_MODE_OFF
, errp
);
5079 /* Downgrade last (so unsupported features can be removed before) */
5080 if (new_version
< old_version
) {
5081 helper_cb_info
.current_operation
= QCOW2_DOWNGRADING
;
5082 ret
= qcow2_downgrade(bs
, new_version
, &qcow2_amend_helper_cb
,
5083 &helper_cb_info
, errp
);
5093 * If offset or size are negative, respectively, they will not be included in
5094 * the BLOCK_IMAGE_CORRUPTED event emitted.
5095 * fatal will be ignored for read-only BDS; corruptions found there will always
5096 * be considered non-fatal.
5098 void qcow2_signal_corruption(BlockDriverState
*bs
, bool fatal
, int64_t offset
,
5099 int64_t size
, const char *message_format
, ...)
5101 BDRVQcow2State
*s
= bs
->opaque
;
5102 const char *node_name
;
5106 fatal
= fatal
&& bdrv_is_writable(bs
);
5108 if (s
->signaled_corruption
&&
5109 (!fatal
|| (s
->incompatible_features
& QCOW2_INCOMPAT_CORRUPT
)))
5114 va_start(ap
, message_format
);
5115 message
= g_strdup_vprintf(message_format
, ap
);
5119 fprintf(stderr
, "qcow2: Marking image as corrupt: %s; further "
5120 "corruption events will be suppressed\n", message
);
5122 fprintf(stderr
, "qcow2: Image is corrupt: %s; further non-fatal "
5123 "corruption events will be suppressed\n", message
);
5126 node_name
= bdrv_get_node_name(bs
);
5127 qapi_event_send_block_image_corrupted(bdrv_get_device_name(bs
),
5128 *node_name
!= '\0', node_name
,
5129 message
, offset
>= 0, offset
,
5135 qcow2_mark_corrupt(bs
);
5136 bs
->drv
= NULL
; /* make BDS unusable */
5139 s
->signaled_corruption
= true;
5142 static QemuOptsList qcow2_create_opts
= {
5143 .name
= "qcow2-create-opts",
5144 .head
= QTAILQ_HEAD_INITIALIZER(qcow2_create_opts
.head
),
5147 .name
= BLOCK_OPT_SIZE
,
5148 .type
= QEMU_OPT_SIZE
,
5149 .help
= "Virtual disk size"
5152 .name
= BLOCK_OPT_COMPAT_LEVEL
,
5153 .type
= QEMU_OPT_STRING
,
5154 .help
= "Compatibility level (0.10 or 1.1)"
5157 .name
= BLOCK_OPT_BACKING_FILE
,
5158 .type
= QEMU_OPT_STRING
,
5159 .help
= "File name of a base image"
5162 .name
= BLOCK_OPT_BACKING_FMT
,
5163 .type
= QEMU_OPT_STRING
,
5164 .help
= "Image format of the base image"
5167 .name
= BLOCK_OPT_DATA_FILE
,
5168 .type
= QEMU_OPT_STRING
,
5169 .help
= "File name of an external data file"
5172 .name
= BLOCK_OPT_DATA_FILE_RAW
,
5173 .type
= QEMU_OPT_BOOL
,
5174 .help
= "The external data file must stay valid as a raw image"
5177 .name
= BLOCK_OPT_ENCRYPT
,
5178 .type
= QEMU_OPT_BOOL
,
5179 .help
= "Encrypt the image with format 'aes'. (Deprecated "
5180 "in favor of " BLOCK_OPT_ENCRYPT_FORMAT
"=aes)",
5183 .name
= BLOCK_OPT_ENCRYPT_FORMAT
,
5184 .type
= QEMU_OPT_STRING
,
5185 .help
= "Encrypt the image, format choices: 'aes', 'luks'",
5187 BLOCK_CRYPTO_OPT_DEF_KEY_SECRET("encrypt.",
5188 "ID of secret providing qcow AES key or LUKS passphrase"),
5189 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_ALG("encrypt."),
5190 BLOCK_CRYPTO_OPT_DEF_LUKS_CIPHER_MODE("encrypt."),
5191 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_ALG("encrypt."),
5192 BLOCK_CRYPTO_OPT_DEF_LUKS_IVGEN_HASH_ALG("encrypt."),
5193 BLOCK_CRYPTO_OPT_DEF_LUKS_HASH_ALG("encrypt."),
5194 BLOCK_CRYPTO_OPT_DEF_LUKS_ITER_TIME("encrypt."),
5196 .name
= BLOCK_OPT_CLUSTER_SIZE
,
5197 .type
= QEMU_OPT_SIZE
,
5198 .help
= "qcow2 cluster size",
5199 .def_value_str
= stringify(DEFAULT_CLUSTER_SIZE
)
5202 .name
= BLOCK_OPT_PREALLOC
,
5203 .type
= QEMU_OPT_STRING
,
5204 .help
= "Preallocation mode (allowed values: off, metadata, "
5208 .name
= BLOCK_OPT_LAZY_REFCOUNTS
,
5209 .type
= QEMU_OPT_BOOL
,
5210 .help
= "Postpone refcount updates",
5211 .def_value_str
= "off"
5214 .name
= BLOCK_OPT_REFCOUNT_BITS
,
5215 .type
= QEMU_OPT_NUMBER
,
5216 .help
= "Width of a reference count entry in bits",
5217 .def_value_str
= "16"
5219 { /* end of list */ }
5223 static const char *const qcow2_strong_runtime_opts
[] = {
5224 "encrypt." BLOCK_CRYPTO_OPT_QCOW_KEY_SECRET
,
5229 BlockDriver bdrv_qcow2
= {
5230 .format_name
= "qcow2",
5231 .instance_size
= sizeof(BDRVQcow2State
),
5232 .bdrv_probe
= qcow2_probe
,
5233 .bdrv_open
= qcow2_open
,
5234 .bdrv_close
= qcow2_close
,
5235 .bdrv_reopen_prepare
= qcow2_reopen_prepare
,
5236 .bdrv_reopen_commit
= qcow2_reopen_commit
,
5237 .bdrv_reopen_abort
= qcow2_reopen_abort
,
5238 .bdrv_join_options
= qcow2_join_options
,
5239 .bdrv_child_perm
= bdrv_format_default_perms
,
5240 .bdrv_co_create_opts
= qcow2_co_create_opts
,
5241 .bdrv_co_create
= qcow2_co_create
,
5242 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
5243 .bdrv_co_block_status
= qcow2_co_block_status
,
5245 .bdrv_co_preadv
= qcow2_co_preadv
,
5246 .bdrv_co_pwritev
= qcow2_co_pwritev
,
5247 .bdrv_co_flush_to_os
= qcow2_co_flush_to_os
,
5249 .bdrv_co_pwrite_zeroes
= qcow2_co_pwrite_zeroes
,
5250 .bdrv_co_pdiscard
= qcow2_co_pdiscard
,
5251 .bdrv_co_copy_range_from
= qcow2_co_copy_range_from
,
5252 .bdrv_co_copy_range_to
= qcow2_co_copy_range_to
,
5253 .bdrv_co_truncate
= qcow2_co_truncate
,
5254 .bdrv_co_pwritev_compressed
= qcow2_co_pwritev_compressed
,
5255 .bdrv_make_empty
= qcow2_make_empty
,
5257 .bdrv_snapshot_create
= qcow2_snapshot_create
,
5258 .bdrv_snapshot_goto
= qcow2_snapshot_goto
,
5259 .bdrv_snapshot_delete
= qcow2_snapshot_delete
,
5260 .bdrv_snapshot_list
= qcow2_snapshot_list
,
5261 .bdrv_snapshot_load_tmp
= qcow2_snapshot_load_tmp
,
5262 .bdrv_measure
= qcow2_measure
,
5263 .bdrv_get_info
= qcow2_get_info
,
5264 .bdrv_get_specific_info
= qcow2_get_specific_info
,
5266 .bdrv_save_vmstate
= qcow2_save_vmstate
,
5267 .bdrv_load_vmstate
= qcow2_load_vmstate
,
5269 .supports_backing
= true,
5270 .bdrv_change_backing_file
= qcow2_change_backing_file
,
5272 .bdrv_refresh_limits
= qcow2_refresh_limits
,
5273 .bdrv_co_invalidate_cache
= qcow2_co_invalidate_cache
,
5274 .bdrv_inactivate
= qcow2_inactivate
,
5276 .create_opts
= &qcow2_create_opts
,
5277 .strong_runtime_opts
= qcow2_strong_runtime_opts
,
5278 .bdrv_co_check
= qcow2_co_check
,
5279 .bdrv_amend_options
= qcow2_amend_options
,
5281 .bdrv_detach_aio_context
= qcow2_detach_aio_context
,
5282 .bdrv_attach_aio_context
= qcow2_attach_aio_context
,
5284 .bdrv_reopen_bitmaps_rw
= qcow2_reopen_bitmaps_rw
,
5285 .bdrv_can_store_new_dirty_bitmap
= qcow2_can_store_new_dirty_bitmap
,
5286 .bdrv_remove_persistent_dirty_bitmap
= qcow2_remove_persistent_dirty_bitmap
,
5289 static void bdrv_qcow2_init(void)
5291 bdrv_register(&bdrv_qcow2
);
5294 block_init(bdrv_qcow2_init
);